Add codeformer and update license

facelib/detection/__init__.py

@@ -0,0 +1,100 @@
+import os
+import torch
+from torch import nn
+from copy import deepcopy
+
+from facelib.utils import load_file_from_url
+from facelib.utils import download_pretrained_models
+from facelib.detection.yolov5face.models.common import Conv
+
+from .retinaface.retinaface import RetinaFace
+from .yolov5face.face_detector import YoloDetector
+
+
+def init_detection_model(model_name, half=False, device='cuda'):
+    if 'retinaface' in model_name:
+        model = init_retinaface_model(model_name, half, device)
+    elif 'YOLOv5' in model_name:
+        model = init_yolov5face_model(model_name, device)
+    else:
+        raise NotImplementedError(f'{model_name} is not implemented.')
+
+    return model
+
+
+def init_retinaface_model(model_name, half=False, device='cuda'):
+    if model_name == 'retinaface_resnet50':
+        model = RetinaFace(network_name='resnet50', half=half)
+        model_url = 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/detection_Resnet50_Final.pth'
+    elif model_name == 'retinaface_mobile0.25':
+        model = RetinaFace(network_name='mobile0.25', half=half)
+        model_url = 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/detection_mobilenet0.25_Final.pth'
+    else:
+        raise NotImplementedError(f'{model_name} is not implemented.')
+
+    model_path = load_file_from_url(url=model_url, model_dir='weights/facelib', progress=True, file_name=None)
+    load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
+    # remove unnecessary 'module.'
+    for k, v in deepcopy(load_net).items():
+        if k.startswith('module.'):
+            load_net[k[7:]] = v
+            load_net.pop(k)
+    model.load_state_dict(load_net, strict=True)
+    model.eval()
+    model = model.to(device)
+
+    return model
+
+
+def init_yolov5face_model(model_name, device='cuda'):
+    if model_name == 'YOLOv5l':
+        model = YoloDetector(config_name='facelib/detection/yolov5face/models/yolov5l.yaml', device=device)
+        model_url = 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/yolov5l-face.pth'
+    elif model_name == 'YOLOv5n':
+        model = YoloDetector(config_name='facelib/detection/yolov5face/models/yolov5n.yaml', device=device)
+        model_url = 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/yolov5n-face.pth'
+    else:
+        raise NotImplementedError(f'{model_name} is not implemented.')
+    
+    model_path = load_file_from_url(url=model_url, model_dir='weights/facelib', progress=True, file_name=None)
+    load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
+    model.detector.load_state_dict(load_net, strict=True)
+    model.detector.eval()
+    model.detector = model.detector.to(device).float()
+
+    for m in model.detector.modules():
+        if type(m) in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
+            m.inplace = True  # pytorch 1.7.0 compatibility
+        elif isinstance(m, Conv):
+            m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatibility
+
+    return model
+
+
+# Download from Google Drive
+# def init_yolov5face_model(model_name, device='cuda'):
+#     if model_name == 'YOLOv5l':
+#         model = YoloDetector(config_name='facelib/detection/yolov5face/models/yolov5l.yaml', device=device)
+#         f_id = {'yolov5l-face.pth': '131578zMA6B2x8VQHyHfa6GEPtulMCNzV'}
+#     elif model_name == 'YOLOv5n':
+#         model = YoloDetector(config_name='facelib/detection/yolov5face/models/yolov5n.yaml', device=device)
+#         f_id = {'yolov5n-face.pth': '1fhcpFvWZqghpGXjYPIne2sw1Fy4yhw6o'}
+#     else:
+#         raise NotImplementedError(f'{model_name} is not implemented.')
+
+#     model_path = os.path.join('weights/facelib', list(f_id.keys())[0])
+#     if not os.path.exists(model_path):
+#         download_pretrained_models(file_ids=f_id, save_path_root='weights/facelib')
+
+#     load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
+#     model.detector.load_state_dict(load_net, strict=True)
+#     model.detector.eval()
+#     model.detector = model.detector.to(device).float()
+
+#     for m in model.detector.modules():
+#         if type(m) in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
+#             m.inplace = True  # pytorch 1.7.0 compatibility
+#         elif isinstance(m, Conv):
+#             m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatibility
+
+#     return model

facelib/detection/align_trans.py

@@ -0,0 +1,219 @@
+import cv2
+import numpy as np
+
+from .matlab_cp2tform import get_similarity_transform_for_cv2
+
+# reference facial points, a list of coordinates (x,y)
+REFERENCE_FACIAL_POINTS = [[30.29459953, 51.69630051], [65.53179932, 51.50139999], [48.02519989, 71.73660278],
+                           [33.54930115, 92.3655014], [62.72990036, 92.20410156]]
+
+DEFAULT_CROP_SIZE = (96, 112)
+
+
+class FaceWarpException(Exception):
+
+    def __str__(self):
+        return 'In File {}:{}'.format(__file__, super.__str__(self))
+
+
+def get_reference_facial_points(output_size=None, inner_padding_factor=0.0, outer_padding=(0, 0), default_square=False):
+    """
+    Function:
+    ----------
+        get reference 5 key points according to crop settings:
+        0. Set default crop_size:
+            if default_square:
+                crop_size = (112, 112)
+            else:
+                crop_size = (96, 112)
+        1. Pad the crop_size by inner_padding_factor in each side;
+        2. Resize crop_size into (output_size - outer_padding*2),
+            pad into output_size with outer_padding;
+        3. Output reference_5point;
+    Parameters:
+    ----------
+        @output_size: (w, h) or None
+            size of aligned face image
+        @inner_padding_factor: (w_factor, h_factor)
+            padding factor for inner (w, h)
+        @outer_padding: (w_pad, h_pad)
+            each row is a pair of coordinates (x, y)
+        @default_square: True or False
+            if True:
+                default crop_size = (112, 112)
+            else:
+                default crop_size = (96, 112);
+        !!! make sure, if output_size is not None:
+                (output_size - outer_padding)
+                = some_scale * (default crop_size * (1.0 +
+                inner_padding_factor))
+    Returns:
+    ----------
+        @reference_5point: 5x2 np.array
+            each row is a pair of transformed coordinates (x, y)
+    """
+
+    tmp_5pts = np.array(REFERENCE_FACIAL_POINTS)
+    tmp_crop_size = np.array(DEFAULT_CROP_SIZE)
+
+    # 0) make the inner region a square
+    if default_square:
+        size_diff = max(tmp_crop_size) - tmp_crop_size
+        tmp_5pts += size_diff / 2
+        tmp_crop_size += size_diff
+
+    if (output_size and output_size[0] == tmp_crop_size[0] and output_size[1] == tmp_crop_size[1]):
+
+        return tmp_5pts
+
+    if (inner_padding_factor == 0 and outer_padding == (0, 0)):
+        if output_size is None:
+            return tmp_5pts
+        else:
+            raise FaceWarpException('No paddings to do, output_size must be None or {}'.format(tmp_crop_size))
+
+    # check output size
+    if not (0 <= inner_padding_factor <= 1.0):
+        raise FaceWarpException('Not (0 <= inner_padding_factor <= 1.0)')
+
+    if ((inner_padding_factor > 0 or outer_padding[0] > 0 or outer_padding[1] > 0) and output_size is None):
+        output_size = tmp_crop_size * \
+            (1 + inner_padding_factor * 2).astype(np.int32)
+        output_size += np.array(outer_padding)
+    if not (outer_padding[0] < output_size[0] and outer_padding[1] < output_size[1]):
+        raise FaceWarpException('Not (outer_padding[0] < output_size[0] and outer_padding[1] < output_size[1])')
+
+    # 1) pad the inner region according inner_padding_factor
+    if inner_padding_factor > 0:
+        size_diff = tmp_crop_size * inner_padding_factor * 2
+        tmp_5pts += size_diff / 2
+        tmp_crop_size += np.round(size_diff).astype(np.int32)
+
+    # 2) resize the padded inner region
+    size_bf_outer_pad = np.array(output_size) - np.array(outer_padding) * 2
+
+    if size_bf_outer_pad[0] * tmp_crop_size[1] != size_bf_outer_pad[1] * tmp_crop_size[0]:
+        raise FaceWarpException('Must have (output_size - outer_padding)'
+                                '= some_scale * (crop_size * (1.0 + inner_padding_factor)')
+
+    scale_factor = size_bf_outer_pad[0].astype(np.float32) / tmp_crop_size[0]
+    tmp_5pts = tmp_5pts * scale_factor
+    #    size_diff = tmp_crop_size * (scale_factor - min(scale_factor))
+    #    tmp_5pts = tmp_5pts + size_diff / 2
+    tmp_crop_size = size_bf_outer_pad
+
+    # 3) add outer_padding to make output_size
+    reference_5point = tmp_5pts + np.array(outer_padding)
+    tmp_crop_size = output_size
+
+    return reference_5point
+
+
+def get_affine_transform_matrix(src_pts, dst_pts):
+    """
+    Function:
+    ----------
+        get affine transform matrix 'tfm' from src_pts to dst_pts
+    Parameters:
+    ----------
+        @src_pts: Kx2 np.array
+            source points matrix, each row is a pair of coordinates (x, y)
+        @dst_pts: Kx2 np.array
+            destination points matrix, each row is a pair of coordinates (x, y)
+    Returns:
+    ----------
+        @tfm: 2x3 np.array
+            transform matrix from src_pts to dst_pts
+    """
+
+    tfm = np.float32([[1, 0, 0], [0, 1, 0]])
+    n_pts = src_pts.shape[0]
+    ones = np.ones((n_pts, 1), src_pts.dtype)
+    src_pts_ = np.hstack([src_pts, ones])
+    dst_pts_ = np.hstack([dst_pts, ones])
+
+    A, res, rank, s = np.linalg.lstsq(src_pts_, dst_pts_)
+
+    if rank == 3:
+        tfm = np.float32([[A[0, 0], A[1, 0], A[2, 0]], [A[0, 1], A[1, 1], A[2, 1]]])
+    elif rank == 2:
+        tfm = np.float32([[A[0, 0], A[1, 0], 0], [A[0, 1], A[1, 1], 0]])
+
+    return tfm
+
+
+def warp_and_crop_face(src_img, facial_pts, reference_pts=None, crop_size=(96, 112), align_type='smilarity'):
+    """
+    Function:
+    ----------
+        apply affine transform 'trans' to uv
+    Parameters:
+    ----------
+        @src_img: 3x3 np.array
+            input image
+        @facial_pts: could be
+            1)a list of K coordinates (x,y)
+        or
+            2) Kx2 or 2xK np.array
+            each row or col is a pair of coordinates (x, y)
+        @reference_pts: could be
+            1) a list of K coordinates (x,y)
+        or
+            2) Kx2 or 2xK np.array
+            each row or col is a pair of coordinates (x, y)
+        or
+            3) None
+            if None, use default reference facial points
+        @crop_size: (w, h)
+            output face image size
+        @align_type: transform type, could be one of
+            1) 'similarity': use similarity transform
+            2) 'cv2_affine': use the first 3 points to do affine transform,
+                    by calling cv2.getAffineTransform()
+            3) 'affine': use all points to do affine transform
+    Returns:
+    ----------
+        @face_img: output face image with size (w, h) = @crop_size
+    """
+
+    if reference_pts is None:
+        if crop_size[0] == 96 and crop_size[1] == 112:
+            reference_pts = REFERENCE_FACIAL_POINTS
+        else:
+            default_square = False
+            inner_padding_factor = 0
+            outer_padding = (0, 0)
+            output_size = crop_size
+
+            reference_pts = get_reference_facial_points(output_size, inner_padding_factor, outer_padding,
+                                                        default_square)
+
+    ref_pts = np.float32(reference_pts)
+    ref_pts_shp = ref_pts.shape
+    if max(ref_pts_shp) < 3 or min(ref_pts_shp) != 2:
+        raise FaceWarpException('reference_pts.shape must be (K,2) or (2,K) and K>2')
+
+    if ref_pts_shp[0] == 2:
+        ref_pts = ref_pts.T
+
+    src_pts = np.float32(facial_pts)
+    src_pts_shp = src_pts.shape
+    if max(src_pts_shp) < 3 or min(src_pts_shp) != 2:
+        raise FaceWarpException('facial_pts.shape must be (K,2) or (2,K) and K>2')
+
+    if src_pts_shp[0] == 2:
+        src_pts = src_pts.T
+
+    if src_pts.shape != ref_pts.shape:
+        raise FaceWarpException('facial_pts and reference_pts must have the same shape')
+
+    if align_type == 'cv2_affine':
+        tfm = cv2.getAffineTransform(src_pts[0:3], ref_pts[0:3])
+    elif align_type == 'affine':
+        tfm = get_affine_transform_matrix(src_pts, ref_pts)
+    else:
+        tfm = get_similarity_transform_for_cv2(src_pts, ref_pts)
+
+    face_img = cv2.warpAffine(src_img, tfm, (crop_size[0], crop_size[1]))
+
+    return face_img

facelib/detection/matlab_cp2tform.py

@@ -0,0 +1,317 @@
+import numpy as np
+from numpy.linalg import inv, lstsq
+from numpy.linalg import matrix_rank as rank
+from numpy.linalg import norm
+
+
+class MatlabCp2tormException(Exception):
+
+    def __str__(self):
+        return 'In File {}:{}'.format(__file__, super.__str__(self))
+
+
+def tformfwd(trans, uv):
+    """
+    Function:
+    ----------
+        apply affine transform 'trans' to uv
+
+    Parameters:
+    ----------
+        @trans: 3x3 np.array
+            transform matrix
+        @uv: Kx2 np.array
+            each row is a pair of coordinates (x, y)
+
+    Returns:
+    ----------
+        @xy: Kx2 np.array
+            each row is a pair of transformed coordinates (x, y)
+    """
+    uv = np.hstack((uv, np.ones((uv.shape[0], 1))))
+    xy = np.dot(uv, trans)
+    xy = xy[:, 0:-1]
+    return xy
+
+
+def tforminv(trans, uv):
+    """
+    Function:
+    ----------
+        apply the inverse of affine transform 'trans' to uv
+
+    Parameters:
+    ----------
+        @trans: 3x3 np.array
+            transform matrix
+        @uv: Kx2 np.array
+            each row is a pair of coordinates (x, y)
+
+    Returns:
+    ----------
+        @xy: Kx2 np.array
+            each row is a pair of inverse-transformed coordinates (x, y)
+    """
+    Tinv = inv(trans)
+    xy = tformfwd(Tinv, uv)
+    return xy
+
+
+def findNonreflectiveSimilarity(uv, xy, options=None):
+    options = {'K': 2}
+
+    K = options['K']
+    M = xy.shape[0]
+    x = xy[:, 0].reshape((-1, 1))  # use reshape to keep a column vector
+    y = xy[:, 1].reshape((-1, 1))  # use reshape to keep a column vector
+
+    tmp1 = np.hstack((x, y, np.ones((M, 1)), np.zeros((M, 1))))
+    tmp2 = np.hstack((y, -x, np.zeros((M, 1)), np.ones((M, 1))))
+    X = np.vstack((tmp1, tmp2))
+
+    u = uv[:, 0].reshape((-1, 1))  # use reshape to keep a column vector
+    v = uv[:, 1].reshape((-1, 1))  # use reshape to keep a column vector
+    U = np.vstack((u, v))
+
+    # We know that X * r = U
+    if rank(X) >= 2 * K:
+        r, _, _, _ = lstsq(X, U, rcond=-1)
+        r = np.squeeze(r)
+    else:
+        raise Exception('cp2tform:twoUniquePointsReq')
+    sc = r[0]
+    ss = r[1]
+    tx = r[2]
+    ty = r[3]
+
+    Tinv = np.array([[sc, -ss, 0], [ss, sc, 0], [tx, ty, 1]])
+    T = inv(Tinv)
+    T[:, 2] = np.array([0, 0, 1])
+
+    return T, Tinv
+
+
+def findSimilarity(uv, xy, options=None):
+    options = {'K': 2}
+
+    #    uv = np.array(uv)
+    #    xy = np.array(xy)
+
+    # Solve for trans1
+    trans1, trans1_inv = findNonreflectiveSimilarity(uv, xy, options)
+
+    # Solve for trans2
+
+    # manually reflect the xy data across the Y-axis
+    xyR = xy
+    xyR[:, 0] = -1 * xyR[:, 0]
+
+    trans2r, trans2r_inv = findNonreflectiveSimilarity(uv, xyR, options)
+
+    # manually reflect the tform to undo the reflection done on xyR
+    TreflectY = np.array([[-1, 0, 0], [0, 1, 0], [0, 0, 1]])
+
+    trans2 = np.dot(trans2r, TreflectY)
+
+    # Figure out if trans1 or trans2 is better
+    xy1 = tformfwd(trans1, uv)
+    norm1 = norm(xy1 - xy)
+
+    xy2 = tformfwd(trans2, uv)
+    norm2 = norm(xy2 - xy)
+
+    if norm1 <= norm2:
+        return trans1, trans1_inv
+    else:
+        trans2_inv = inv(trans2)
+        return trans2, trans2_inv
+
+
+def get_similarity_transform(src_pts, dst_pts, reflective=True):
+    """
+    Function:
+    ----------
+        Find Similarity Transform Matrix 'trans':
+            u = src_pts[:, 0]
+            v = src_pts[:, 1]
+            x = dst_pts[:, 0]
+            y = dst_pts[:, 1]
+            [x, y, 1] = [u, v, 1] * trans
+
+    Parameters:
+    ----------
+        @src_pts: Kx2 np.array
+            source points, each row is a pair of coordinates (x, y)
+        @dst_pts: Kx2 np.array
+            destination points, each row is a pair of transformed
+            coordinates (x, y)
+        @reflective: True or False
+            if True:
+                use reflective similarity transform
+            else:
+                use non-reflective similarity transform
+
+    Returns:
+    ----------
+       @trans: 3x3 np.array
+            transform matrix from uv to xy
+        trans_inv: 3x3 np.array
+            inverse of trans, transform matrix from xy to uv
+    """
+
+    if reflective:
+        trans, trans_inv = findSimilarity(src_pts, dst_pts)
+    else:
+        trans, trans_inv = findNonreflectiveSimilarity(src_pts, dst_pts)
+
+    return trans, trans_inv
+
+
+def cvt_tform_mat_for_cv2(trans):
+    """
+    Function:
+    ----------
+        Convert Transform Matrix 'trans' into 'cv2_trans' which could be
+        directly used by cv2.warpAffine():
+            u = src_pts[:, 0]
+            v = src_pts[:, 1]
+            x = dst_pts[:, 0]
+            y = dst_pts[:, 1]
+            [x, y].T = cv_trans * [u, v, 1].T
+
+    Parameters:
+    ----------
+        @trans: 3x3 np.array
+            transform matrix from uv to xy
+
+    Returns:
+    ----------
+        @cv2_trans: 2x3 np.array
+            transform matrix from src_pts to dst_pts, could be directly used
+            for cv2.warpAffine()
+    """
+    cv2_trans = trans[:, 0:2].T
+
+    return cv2_trans
+
+
+def get_similarity_transform_for_cv2(src_pts, dst_pts, reflective=True):
+    """
+    Function:
+    ----------
+        Find Similarity Transform Matrix 'cv2_trans' which could be
+        directly used by cv2.warpAffine():
+            u = src_pts[:, 0]
+            v = src_pts[:, 1]
+            x = dst_pts[:, 0]
+            y = dst_pts[:, 1]
+            [x, y].T = cv_trans * [u, v, 1].T
+
+    Parameters:
+    ----------
+        @src_pts: Kx2 np.array
+            source points, each row is a pair of coordinates (x, y)
+        @dst_pts: Kx2 np.array
+            destination points, each row is a pair of transformed
+            coordinates (x, y)
+        reflective: True or False
+            if True:
+                use reflective similarity transform
+            else:
+                use non-reflective similarity transform
+
+    Returns:
+    ----------
+        @cv2_trans: 2x3 np.array
+            transform matrix from src_pts to dst_pts, could be directly used
+            for cv2.warpAffine()
+    """
+    trans, trans_inv = get_similarity_transform(src_pts, dst_pts, reflective)
+    cv2_trans = cvt_tform_mat_for_cv2(trans)
+
+    return cv2_trans
+
+
+if __name__ == '__main__':
+    """
+    u = [0, 6, -2]
+    v = [0, 3, 5]
+    x = [-1, 0, 4]
+    y = [-1, -10, 4]
+
+    # In Matlab, run:
+    #
+    #   uv = [u'; v'];
+    #   xy = [x'; y'];
+    #   tform_sim=cp2tform(uv,xy,'similarity');
+    #
+    #   trans = tform_sim.tdata.T
+    #   ans =
+    #       -0.0764   -1.6190         0
+    #        1.6190   -0.0764         0
+    #       -3.2156    0.0290    1.0000
+    #   trans_inv = tform_sim.tdata.Tinv
+    #    ans =
+    #
+    #       -0.0291    0.6163         0
+    #       -0.6163   -0.0291         0
+    #       -0.0756    1.9826    1.0000
+    #    xy_m=tformfwd(tform_sim, u,v)
+    #
+    #    xy_m =
+    #
+    #       -3.2156    0.0290
+    #        1.1833   -9.9143
+    #        5.0323    2.8853
+    #    uv_m=tforminv(tform_sim, x,y)
+    #
+    #    uv_m =
+    #
+    #        0.5698    1.3953
+    #        6.0872    2.2733
+    #       -2.6570    4.3314
+    """
+    u = [0, 6, -2]
+    v = [0, 3, 5]
+    x = [-1, 0, 4]
+    y = [-1, -10, 4]
+
+    uv = np.array((u, v)).T
+    xy = np.array((x, y)).T
+
+    print('\n--->uv:')
+    print(uv)
+    print('\n--->xy:')
+    print(xy)
+
+    trans, trans_inv = get_similarity_transform(uv, xy)
+
+    print('\n--->trans matrix:')
+    print(trans)
+
+    print('\n--->trans_inv matrix:')
+    print(trans_inv)
+
+    print('\n---> apply transform to uv')
+    print('\nxy_m = uv_augmented * trans')
+    uv_aug = np.hstack((uv, np.ones((uv.shape[0], 1))))
+    xy_m = np.dot(uv_aug, trans)
+    print(xy_m)
+
+    print('\nxy_m = tformfwd(trans, uv)')
+    xy_m = tformfwd(trans, uv)
+    print(xy_m)
+
+    print('\n---> apply inverse transform to xy')
+    print('\nuv_m = xy_augmented * trans_inv')
+    xy_aug = np.hstack((xy, np.ones((xy.shape[0], 1))))
+    uv_m = np.dot(xy_aug, trans_inv)
+    print(uv_m)
+
+    print('\nuv_m = tformfwd(trans_inv, xy)')
+    uv_m = tformfwd(trans_inv, xy)
+    print(uv_m)
+
+    uv_m = tforminv(trans, xy)
+    print('\nuv_m = tforminv(trans, xy)')
+    print(uv_m)

facelib/detection/retinaface/retinaface.py

@@ -0,0 +1,370 @@
+import cv2
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from PIL import Image
+from torchvision.models._utils import IntermediateLayerGetter as IntermediateLayerGetter
+
+from facelib.detection.align_trans import get_reference_facial_points, warp_and_crop_face
+from facelib.detection.retinaface.retinaface_net import FPN, SSH, MobileNetV1, make_bbox_head, make_class_head, make_landmark_head
+from facelib.detection.retinaface.retinaface_utils import (PriorBox, batched_decode, batched_decode_landm, decode, decode_landm,
+                                                 py_cpu_nms)
+
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+
+def generate_config(network_name):
+
+    cfg_mnet = {
+        'name': 'mobilenet0.25',
+        'min_sizes': [[16, 32], [64, 128], [256, 512]],
+        'steps': [8, 16, 32],
+        'variance': [0.1, 0.2],
+        'clip': False,
+        'loc_weight': 2.0,
+        'gpu_train': True,
+        'batch_size': 32,
+        'ngpu': 1,
+        'epoch': 250,
+        'decay1': 190,
+        'decay2': 220,
+        'image_size': 640,
+        'return_layers': {
+            'stage1': 1,
+            'stage2': 2,
+            'stage3': 3
+        },
+        'in_channel': 32,
+        'out_channel': 64
+    }
+
+    cfg_re50 = {
+        'name': 'Resnet50',
+        'min_sizes': [[16, 32], [64, 128], [256, 512]],
+        'steps': [8, 16, 32],
+        'variance': [0.1, 0.2],
+        'clip': False,
+        'loc_weight': 2.0,
+        'gpu_train': True,
+        'batch_size': 24,
+        'ngpu': 4,
+        'epoch': 100,
+        'decay1': 70,
+        'decay2': 90,
+        'image_size': 840,
+        'return_layers': {
+            'layer2': 1,
+            'layer3': 2,
+            'layer4': 3
+        },
+        'in_channel': 256,
+        'out_channel': 256
+    }
+
+    if network_name == 'mobile0.25':
+        return cfg_mnet
+    elif network_name == 'resnet50':
+        return cfg_re50
+    else:
+        raise NotImplementedError(f'network_name={network_name}')
+
+
+class RetinaFace(nn.Module):
+
+    def __init__(self, network_name='resnet50', half=False, phase='test'):
+        super(RetinaFace, self).__init__()
+        self.half_inference = half
+        cfg = generate_config(network_name)
+        self.backbone = cfg['name']
+
+        self.model_name = f'retinaface_{network_name}'
+        self.cfg = cfg
+        self.phase = phase
+        self.target_size, self.max_size = 1600, 2150
+        self.resize, self.scale, self.scale1 = 1., None, None
+        self.mean_tensor = torch.tensor([[[[104.]], [[117.]], [[123.]]]]).to(device)
+        self.reference = get_reference_facial_points(default_square=True)
+        # Build network.
+        backbone = None
+        if cfg['name'] == 'mobilenet0.25':
+            backbone = MobileNetV1()
+            self.body = IntermediateLayerGetter(backbone, cfg['return_layers'])
+        elif cfg['name'] == 'Resnet50':
+            import torchvision.models as models
+            backbone = models.resnet50(pretrained=False)
+            self.body = IntermediateLayerGetter(backbone, cfg['return_layers'])
+
+        in_channels_stage2 = cfg['in_channel']
+        in_channels_list = [
+            in_channels_stage2 * 2,
+            in_channels_stage2 * 4,
+            in_channels_stage2 * 8,
+        ]
+
+        out_channels = cfg['out_channel']
+        self.fpn = FPN(in_channels_list, out_channels)
+        self.ssh1 = SSH(out_channels, out_channels)
+        self.ssh2 = SSH(out_channels, out_channels)
+        self.ssh3 = SSH(out_channels, out_channels)
+
+        self.ClassHead = make_class_head(fpn_num=3, inchannels=cfg['out_channel'])
+        self.BboxHead = make_bbox_head(fpn_num=3, inchannels=cfg['out_channel'])
+        self.LandmarkHead = make_landmark_head(fpn_num=3, inchannels=cfg['out_channel'])
+
+        self.to(device)
+        self.eval()
+        if self.half_inference:
+            self.half()
+
+    def forward(self, inputs):
+        out = self.body(inputs)
+
+        if self.backbone == 'mobilenet0.25' or self.backbone == 'Resnet50':
+            out = list(out.values())
+        # FPN
+        fpn = self.fpn(out)
+
+        # SSH
+        feature1 = self.ssh1(fpn[0])
+        feature2 = self.ssh2(fpn[1])
+        feature3 = self.ssh3(fpn[2])
+        features = [feature1, feature2, feature3]
+
+        bbox_regressions = torch.cat([self.BboxHead[i](feature) for i, feature in enumerate(features)], dim=1)
+        classifications = torch.cat([self.ClassHead[i](feature) for i, feature in enumerate(features)], dim=1)
+        tmp = [self.LandmarkHead[i](feature) for i, feature in enumerate(features)]
+        ldm_regressions = (torch.cat(tmp, dim=1))
+
+        if self.phase == 'train':
+            output = (bbox_regressions, classifications, ldm_regressions)
+        else:
+            output = (bbox_regressions, F.softmax(classifications, dim=-1), ldm_regressions)
+        return output
+
+    def __detect_faces(self, inputs):
+        # get scale
+        height, width = inputs.shape[2:]
+        self.scale = torch.tensor([width, height, width, height], dtype=torch.float32).to(device)
+        tmp = [width, height, width, height, width, height, width, height, width, height]
+        self.scale1 = torch.tensor(tmp, dtype=torch.float32).to(device)
+
+        # forawrd
+        inputs = inputs.to(device)
+        if self.half_inference:
+            inputs = inputs.half()
+        loc, conf, landmarks = self(inputs)
+
+        # get priorbox
+        priorbox = PriorBox(self.cfg, image_size=inputs.shape[2:])
+        priors = priorbox.forward().to(device)
+
+        return loc, conf, landmarks, priors
+
+    # single image detection
+    def transform(self, image, use_origin_size):
+        # convert to opencv format
+        if isinstance(image, Image.Image):
+            image = cv2.cvtColor(np.asarray(image), cv2.COLOR_RGB2BGR)
+        image = image.astype(np.float32)
+
+        # testing scale
+        im_size_min = np.min(image.shape[0:2])
+        im_size_max = np.max(image.shape[0:2])
+        resize = float(self.target_size) / float(im_size_min)
+
+        # prevent bigger axis from being more than max_size
+        if np.round(resize * im_size_max) > self.max_size:
+            resize = float(self.max_size) / float(im_size_max)
+        resize = 1 if use_origin_size else resize
+
+        # resize
+        if resize != 1:
+            image = cv2.resize(image, None, None, fx=resize, fy=resize, interpolation=cv2.INTER_LINEAR)
+
+        # convert to torch.tensor format
+        # image -= (104, 117, 123)
+        image = image.transpose(2, 0, 1)
+        image = torch.from_numpy(image).unsqueeze(0)
+
+        return image, resize
+
+    def detect_faces(
+        self,
+        image,
+        conf_threshold=0.8,
+        nms_threshold=0.4,
+        use_origin_size=True,
+    ):
+        """
+        Params:
+            imgs: BGR image
+        """
+        image, self.resize = self.transform(image, use_origin_size)
+        image = image.to(device)
+        if self.half_inference:
+            image = image.half()
+        image = image - self.mean_tensor
+
+        loc, conf, landmarks, priors = self.__detect_faces(image)
+
+        boxes = decode(loc.data.squeeze(0), priors.data, self.cfg['variance'])
+        boxes = boxes * self.scale / self.resize
+        boxes = boxes.cpu().numpy()
+
+        scores = conf.squeeze(0).data.cpu().numpy()[:, 1]
+
+        landmarks = decode_landm(landmarks.squeeze(0), priors, self.cfg['variance'])
+        landmarks = landmarks * self.scale1 / self.resize
+        landmarks = landmarks.cpu().numpy()
+
+        # ignore low scores
+        inds = np.where(scores > conf_threshold)[0]
+        boxes, landmarks, scores = boxes[inds], landmarks[inds], scores[inds]
+
+        # sort
+        order = scores.argsort()[::-1]
+        boxes, landmarks, scores = boxes[order], landmarks[order], scores[order]
+
+        # do NMS
+        bounding_boxes = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False)
+        keep = py_cpu_nms(bounding_boxes, nms_threshold)
+        bounding_boxes, landmarks = bounding_boxes[keep, :], landmarks[keep]
+        # self.t['forward_pass'].toc()
+        # print(self.t['forward_pass'].average_time)
+        # import sys
+        # sys.stdout.flush()
+        return np.concatenate((bounding_boxes, landmarks), axis=1)
+
+    def __align_multi(self, image, boxes, landmarks, limit=None):
+
+        if len(boxes) < 1:
+            return [], []
+
+        if limit:
+            boxes = boxes[:limit]
+            landmarks = landmarks[:limit]
+
+        faces = []
+        for landmark in landmarks:
+            facial5points = [[landmark[2 * j], landmark[2 * j + 1]] for j in range(5)]
+
+            warped_face = warp_and_crop_face(np.array(image), facial5points, self.reference, crop_size=(112, 112))
+            faces.append(warped_face)
+
+        return np.concatenate((boxes, landmarks), axis=1), faces
+
+    def align_multi(self, img, conf_threshold=0.8, limit=None):
+
+        rlt = self.detect_faces(img, conf_threshold=conf_threshold)
+        boxes, landmarks = rlt[:, 0:5], rlt[:, 5:]
+
+        return self.__align_multi(img, boxes, landmarks, limit)
+
+    # batched detection
+    def batched_transform(self, frames, use_origin_size):
+        """
+        Arguments:
+            frames: a list of PIL.Image, or torch.Tensor(shape=[n, h, w, c],
+                type=np.float32, BGR format).
+            use_origin_size: whether to use origin size.
+        """
+        from_PIL = True if isinstance(frames[0], Image.Image) else False
+
+        # convert to opencv format
+        if from_PIL:
+            frames = [cv2.cvtColor(np.asarray(frame), cv2.COLOR_RGB2BGR) for frame in frames]
+            frames = np.asarray(frames, dtype=np.float32)
+
+        # testing scale
+        im_size_min = np.min(frames[0].shape[0:2])
+        im_size_max = np.max(frames[0].shape[0:2])
+        resize = float(self.target_size) / float(im_size_min)
+
+        # prevent bigger axis from being more than max_size
+        if np.round(resize * im_size_max) > self.max_size:
+            resize = float(self.max_size) / float(im_size_max)
+        resize = 1 if use_origin_size else resize
+
+        # resize
+        if resize != 1:
+            if not from_PIL:
+                frames = F.interpolate(frames, scale_factor=resize)
+            else:
+                frames = [
+                    cv2.resize(frame, None, None, fx=resize, fy=resize, interpolation=cv2.INTER_LINEAR)
+                    for frame in frames
+                ]
+
+        # convert to torch.tensor format
+        if not from_PIL:
+            frames = frames.transpose(1, 2).transpose(1, 3).contiguous()
+        else:
+            frames = frames.transpose((0, 3, 1, 2))
+            frames = torch.from_numpy(frames)
+
+        return frames, resize
+
+    def batched_detect_faces(self, frames, conf_threshold=0.8, nms_threshold=0.4, use_origin_size=True):
+        """
+        Arguments:
+            frames: a list of PIL.Image, or np.array(shape=[n, h, w, c],
+                type=np.uint8, BGR format).
+            conf_threshold: confidence threshold.
+            nms_threshold: nms threshold.
+            use_origin_size: whether to use origin size.
+        Returns:
+            final_bounding_boxes: list of np.array ([n_boxes, 5],
+                type=np.float32).
+            final_landmarks: list of np.array ([n_boxes, 10], type=np.float32).
+        """
+        # self.t['forward_pass'].tic()
+        frames, self.resize = self.batched_transform(frames, use_origin_size)
+        frames = frames.to(device)
+        frames = frames - self.mean_tensor
+
+        b_loc, b_conf, b_landmarks, priors = self.__detect_faces(frames)
+
+        final_bounding_boxes, final_landmarks = [], []
+
+        # decode
+        priors = priors.unsqueeze(0)
+        b_loc = batched_decode(b_loc, priors, self.cfg['variance']) * self.scale / self.resize
+        b_landmarks = batched_decode_landm(b_landmarks, priors, self.cfg['variance']) * self.scale1 / self.resize
+        b_conf = b_conf[:, :, 1]
+
+        # index for selection
+        b_indice = b_conf > conf_threshold
+
+        # concat
+        b_loc_and_conf = torch.cat((b_loc, b_conf.unsqueeze(-1)), dim=2).float()
+
+        for pred, landm, inds in zip(b_loc_and_conf, b_landmarks, b_indice):
+
+            # ignore low scores
+            pred, landm = pred[inds, :], landm[inds, :]
+            if pred.shape[0] == 0:
+                final_bounding_boxes.append(np.array([], dtype=np.float32))
+                final_landmarks.append(np.array([], dtype=np.float32))
+                continue
+
+            # sort
+            # order = score.argsort(descending=True)
+            # box, landm, score = box[order], landm[order], score[order]
+
+            # to CPU
+            bounding_boxes, landm = pred.cpu().numpy(), landm.cpu().numpy()
+
+            # NMS
+            keep = py_cpu_nms(bounding_boxes, nms_threshold)
+            bounding_boxes, landmarks = bounding_boxes[keep, :], landm[keep]
+
+            # append
+            final_bounding_boxes.append(bounding_boxes)
+            final_landmarks.append(landmarks)
+        # self.t['forward_pass'].toc(average=True)
+        # self.batch_time += self.t['forward_pass'].diff
+        # self.total_frame += len(frames)
+        # print(self.batch_time / self.total_frame)
+
+        return final_bounding_boxes, final_landmarks

facelib/detection/retinaface/retinaface_net.py

@@ -0,0 +1,196 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+def conv_bn(inp, oup, stride=1, leaky=0):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 3, stride, 1, bias=False), nn.BatchNorm2d(oup),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True))
+
+
+def conv_bn_no_relu(inp, oup, stride):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
+        nn.BatchNorm2d(oup),
+    )
+
+
+def conv_bn1X1(inp, oup, stride, leaky=0):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 1, stride, padding=0, bias=False), nn.BatchNorm2d(oup),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True))
+
+
+def conv_dw(inp, oup, stride, leaky=0.1):
+    return nn.Sequential(
+        nn.Conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
+        nn.BatchNorm2d(inp),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True),
+        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
+        nn.BatchNorm2d(oup),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True),
+    )
+
+
+class SSH(nn.Module):
+
+    def __init__(self, in_channel, out_channel):
+        super(SSH, self).__init__()
+        assert out_channel % 4 == 0
+        leaky = 0
+        if (out_channel <= 64):
+            leaky = 0.1
+        self.conv3X3 = conv_bn_no_relu(in_channel, out_channel // 2, stride=1)
+
+        self.conv5X5_1 = conv_bn(in_channel, out_channel // 4, stride=1, leaky=leaky)
+        self.conv5X5_2 = conv_bn_no_relu(out_channel // 4, out_channel // 4, stride=1)
+
+        self.conv7X7_2 = conv_bn(out_channel // 4, out_channel // 4, stride=1, leaky=leaky)
+        self.conv7x7_3 = conv_bn_no_relu(out_channel // 4, out_channel // 4, stride=1)
+
+    def forward(self, input):
+        conv3X3 = self.conv3X3(input)
+
+        conv5X5_1 = self.conv5X5_1(input)
+        conv5X5 = self.conv5X5_2(conv5X5_1)
+
+        conv7X7_2 = self.conv7X7_2(conv5X5_1)
+        conv7X7 = self.conv7x7_3(conv7X7_2)
+
+        out = torch.cat([conv3X3, conv5X5, conv7X7], dim=1)
+        out = F.relu(out)
+        return out
+
+
+class FPN(nn.Module):
+
+    def __init__(self, in_channels_list, out_channels):
+        super(FPN, self).__init__()
+        leaky = 0
+        if (out_channels <= 64):
+            leaky = 0.1
+        self.output1 = conv_bn1X1(in_channels_list[0], out_channels, stride=1, leaky=leaky)
+        self.output2 = conv_bn1X1(in_channels_list[1], out_channels, stride=1, leaky=leaky)
+        self.output3 = conv_bn1X1(in_channels_list[2], out_channels, stride=1, leaky=leaky)
+
+        self.merge1 = conv_bn(out_channels, out_channels, leaky=leaky)
+        self.merge2 = conv_bn(out_channels, out_channels, leaky=leaky)
+
+    def forward(self, input):
+        # names = list(input.keys())
+        # input = list(input.values())
+
+        output1 = self.output1(input[0])
+        output2 = self.output2(input[1])
+        output3 = self.output3(input[2])
+
+        up3 = F.interpolate(output3, size=[output2.size(2), output2.size(3)], mode='nearest')
+        output2 = output2 + up3
+        output2 = self.merge2(output2)
+
+        up2 = F.interpolate(output2, size=[output1.size(2), output1.size(3)], mode='nearest')
+        output1 = output1 + up2
+        output1 = self.merge1(output1)
+
+        out = [output1, output2, output3]
+        return out
+
+
+class MobileNetV1(nn.Module):
+
+    def __init__(self):
+        super(MobileNetV1, self).__init__()
+        self.stage1 = nn.Sequential(
+            conv_bn(3, 8, 2, leaky=0.1),  # 3
+            conv_dw(8, 16, 1),  # 7
+            conv_dw(16, 32, 2),  # 11
+            conv_dw(32, 32, 1),  # 19
+            conv_dw(32, 64, 2),  # 27
+            conv_dw(64, 64, 1),  # 43
+        )
+        self.stage2 = nn.Sequential(
+            conv_dw(64, 128, 2),  # 43 + 16 = 59
+            conv_dw(128, 128, 1),  # 59 + 32 = 91
+            conv_dw(128, 128, 1),  # 91 + 32 = 123
+            conv_dw(128, 128, 1),  # 123 + 32 = 155
+            conv_dw(128, 128, 1),  # 155 + 32 = 187
+            conv_dw(128, 128, 1),  # 187 + 32 = 219
+        )
+        self.stage3 = nn.Sequential(
+            conv_dw(128, 256, 2),  # 219 +3 2 = 241
+            conv_dw(256, 256, 1),  # 241 + 64 = 301
+        )
+        self.avg = nn.AdaptiveAvgPool2d((1, 1))
+        self.fc = nn.Linear(256, 1000)
+
+    def forward(self, x):
+        x = self.stage1(x)
+        x = self.stage2(x)
+        x = self.stage3(x)
+        x = self.avg(x)
+        # x = self.model(x)
+        x = x.view(-1, 256)
+        x = self.fc(x)
+        return x
+
+
+class ClassHead(nn.Module):
+
+    def __init__(self, inchannels=512, num_anchors=3):
+        super(ClassHead, self).__init__()
+        self.num_anchors = num_anchors
+        self.conv1x1 = nn.Conv2d(inchannels, self.num_anchors * 2, kernel_size=(1, 1), stride=1, padding=0)
+
+    def forward(self, x):
+        out = self.conv1x1(x)
+        out = out.permute(0, 2, 3, 1).contiguous()
+
+        return out.view(out.shape[0], -1, 2)
+
+
+class BboxHead(nn.Module):
+
+    def __init__(self, inchannels=512, num_anchors=3):
+        super(BboxHead, self).__init__()
+        self.conv1x1 = nn.Conv2d(inchannels, num_anchors * 4, kernel_size=(1, 1), stride=1, padding=0)
+
+    def forward(self, x):
+        out = self.conv1x1(x)
+        out = out.permute(0, 2, 3, 1).contiguous()
+
+        return out.view(out.shape[0], -1, 4)
+
+
+class LandmarkHead(nn.Module):
+
+    def __init__(self, inchannels=512, num_anchors=3):
+        super(LandmarkHead, self).__init__()
+        self.conv1x1 = nn.Conv2d(inchannels, num_anchors * 10, kernel_size=(1, 1), stride=1, padding=0)
+
+    def forward(self, x):
+        out = self.conv1x1(x)
+        out = out.permute(0, 2, 3, 1).contiguous()
+
+        return out.view(out.shape[0], -1, 10)
+
+
+def make_class_head(fpn_num=3, inchannels=64, anchor_num=2):
+    classhead = nn.ModuleList()
+    for i in range(fpn_num):
+        classhead.append(ClassHead(inchannels, anchor_num))
+    return classhead
+
+
+def make_bbox_head(fpn_num=3, inchannels=64, anchor_num=2):
+    bboxhead = nn.ModuleList()
+    for i in range(fpn_num):
+        bboxhead.append(BboxHead(inchannels, anchor_num))
+    return bboxhead
+
+
+def make_landmark_head(fpn_num=3, inchannels=64, anchor_num=2):
+    landmarkhead = nn.ModuleList()
+    for i in range(fpn_num):
+        landmarkhead.append(LandmarkHead(inchannels, anchor_num))
+    return landmarkhead

facelib/detection/retinaface/retinaface_utils.py

@@ -0,0 +1,421 @@
+import numpy as np
+import torch
+import torchvision
+from itertools import product as product
+from math import ceil
+
+
+class PriorBox(object):
+
+    def __init__(self, cfg, image_size=None, phase='train'):
+        super(PriorBox, self).__init__()
+        self.min_sizes = cfg['min_sizes']
+        self.steps = cfg['steps']
+        self.clip = cfg['clip']
+        self.image_size = image_size
+        self.feature_maps = [[ceil(self.image_size[0] / step), ceil(self.image_size[1] / step)] for step in self.steps]
+        self.name = 's'
+
+    def forward(self):
+        anchors = []
+        for k, f in enumerate(self.feature_maps):
+            min_sizes = self.min_sizes[k]
+            for i, j in product(range(f[0]), range(f[1])):
+                for min_size in min_sizes:
+                    s_kx = min_size / self.image_size[1]
+                    s_ky = min_size / self.image_size[0]
+                    dense_cx = [x * self.steps[k] / self.image_size[1] for x in [j + 0.5]]
+                    dense_cy = [y * self.steps[k] / self.image_size[0] for y in [i + 0.5]]
+                    for cy, cx in product(dense_cy, dense_cx):
+                        anchors += [cx, cy, s_kx, s_ky]
+
+        # back to torch land
+        output = torch.Tensor(anchors).view(-1, 4)
+        if self.clip:
+            output.clamp_(max=1, min=0)
+        return output
+
+
+def py_cpu_nms(dets, thresh):
+    """Pure Python NMS baseline."""
+    keep = torchvision.ops.nms(
+        boxes=torch.Tensor(dets[:, :4]),
+        scores=torch.Tensor(dets[:, 4]),
+        iou_threshold=thresh,
+    )
+
+    return list(keep)
+
+
+def point_form(boxes):
+    """ Convert prior_boxes to (xmin, ymin, xmax, ymax)
+    representation for comparison to point form ground truth data.
+    Args:
+        boxes: (tensor) center-size default boxes from priorbox layers.
+    Return:
+        boxes: (tensor) Converted xmin, ymin, xmax, ymax form of boxes.
+    """
+    return torch.cat(
+        (
+            boxes[:, :2] - boxes[:, 2:] / 2,  # xmin, ymin
+            boxes[:, :2] + boxes[:, 2:] / 2),
+        1)  # xmax, ymax
+
+
+def center_size(boxes):
+    """ Convert prior_boxes to (cx, cy, w, h)
+    representation for comparison to center-size form ground truth data.
+    Args:
+        boxes: (tensor) point_form boxes
+    Return:
+        boxes: (tensor) Converted xmin, ymin, xmax, ymax form of boxes.
+    """
+    return torch.cat(
+        (boxes[:, 2:] + boxes[:, :2]) / 2,  # cx, cy
+        boxes[:, 2:] - boxes[:, :2],
+        1)  # w, h
+
+
+def intersect(box_a, box_b):
+    """ We resize both tensors to [A,B,2] without new malloc:
+    [A,2] -> [A,1,2] -> [A,B,2]
+    [B,2] -> [1,B,2] -> [A,B,2]
+    Then we compute the area of intersect between box_a and box_b.
+    Args:
+      box_a: (tensor) bounding boxes, Shape: [A,4].
+      box_b: (tensor) bounding boxes, Shape: [B,4].
+    Return:
+      (tensor) intersection area, Shape: [A,B].
+    """
+    A = box_a.size(0)
+    B = box_b.size(0)
+    max_xy = torch.min(box_a[:, 2:].unsqueeze(1).expand(A, B, 2), box_b[:, 2:].unsqueeze(0).expand(A, B, 2))
+    min_xy = torch.max(box_a[:, :2].unsqueeze(1).expand(A, B, 2), box_b[:, :2].unsqueeze(0).expand(A, B, 2))
+    inter = torch.clamp((max_xy - min_xy), min=0)
+    return inter[:, :, 0] * inter[:, :, 1]
+
+
+def jaccard(box_a, box_b):
+    """Compute the jaccard overlap of two sets of boxes.  The jaccard overlap
+    is simply the intersection over union of two boxes.  Here we operate on
+    ground truth boxes and default boxes.
+    E.g.:
+        A ∩ B / A ∪ B = A ∩ B / (area(A) + area(B) - A ∩ B)
+    Args:
+        box_a: (tensor) Ground truth bounding boxes, Shape: [num_objects,4]
+        box_b: (tensor) Prior boxes from priorbox layers, Shape: [num_priors,4]
+    Return:
+        jaccard overlap: (tensor) Shape: [box_a.size(0), box_b.size(0)]
+    """
+    inter = intersect(box_a, box_b)
+    area_a = ((box_a[:, 2] - box_a[:, 0]) * (box_a[:, 3] - box_a[:, 1])).unsqueeze(1).expand_as(inter)  # [A,B]
+    area_b = ((box_b[:, 2] - box_b[:, 0]) * (box_b[:, 3] - box_b[:, 1])).unsqueeze(0).expand_as(inter)  # [A,B]
+    union = area_a + area_b - inter
+    return inter / union  # [A,B]
+
+
+def matrix_iou(a, b):
+    """
+    return iou of a and b, numpy version for data augenmentation
+    """
+    lt = np.maximum(a[:, np.newaxis, :2], b[:, :2])
+    rb = np.minimum(a[:, np.newaxis, 2:], b[:, 2:])
+
+    area_i = np.prod(rb - lt, axis=2) * (lt < rb).all(axis=2)
+    area_a = np.prod(a[:, 2:] - a[:, :2], axis=1)
+    area_b = np.prod(b[:, 2:] - b[:, :2], axis=1)
+    return area_i / (area_a[:, np.newaxis] + area_b - area_i)
+
+
+def matrix_iof(a, b):
+    """
+    return iof of a and b, numpy version for data augenmentation
+    """
+    lt = np.maximum(a[:, np.newaxis, :2], b[:, :2])
+    rb = np.minimum(a[:, np.newaxis, 2:], b[:, 2:])
+
+    area_i = np.prod(rb - lt, axis=2) * (lt < rb).all(axis=2)
+    area_a = np.prod(a[:, 2:] - a[:, :2], axis=1)
+    return area_i / np.maximum(area_a[:, np.newaxis], 1)
+
+
+def match(threshold, truths, priors, variances, labels, landms, loc_t, conf_t, landm_t, idx):
+    """Match each prior box with the ground truth box of the highest jaccard
+    overlap, encode the bounding boxes, then return the matched indices
+    corresponding to both confidence and location preds.
+    Args:
+        threshold: (float) The overlap threshold used when matching boxes.
+        truths: (tensor) Ground truth boxes, Shape: [num_obj, 4].
+        priors: (tensor) Prior boxes from priorbox layers, Shape: [n_priors,4].
+        variances: (tensor) Variances corresponding to each prior coord,
+            Shape: [num_priors, 4].
+        labels: (tensor) All the class labels for the image, Shape: [num_obj].
+        landms: (tensor) Ground truth landms, Shape [num_obj, 10].
+        loc_t: (tensor) Tensor to be filled w/ encoded location targets.
+        conf_t: (tensor) Tensor to be filled w/ matched indices for conf preds.
+        landm_t: (tensor) Tensor to be filled w/ encoded landm targets.
+        idx: (int) current batch index
+    Return:
+        The matched indices corresponding to 1)location 2)confidence
+        3)landm preds.
+    """
+    # jaccard index
+    overlaps = jaccard(truths, point_form(priors))
+    # (Bipartite Matching)
+    # [1,num_objects] best prior for each ground truth
+    best_prior_overlap, best_prior_idx = overlaps.max(1, keepdim=True)
+
+    # ignore hard gt
+    valid_gt_idx = best_prior_overlap[:, 0] >= 0.2
+    best_prior_idx_filter = best_prior_idx[valid_gt_idx, :]
+    if best_prior_idx_filter.shape[0] <= 0:
+        loc_t[idx] = 0
+        conf_t[idx] = 0
+        return
+
+    # [1,num_priors] best ground truth for each prior
+    best_truth_overlap, best_truth_idx = overlaps.max(0, keepdim=True)
+    best_truth_idx.squeeze_(0)
+    best_truth_overlap.squeeze_(0)
+    best_prior_idx.squeeze_(1)
+    best_prior_idx_filter.squeeze_(1)
+    best_prior_overlap.squeeze_(1)
+    best_truth_overlap.index_fill_(0, best_prior_idx_filter, 2)  # ensure best prior
+    # TODO refactor: index  best_prior_idx with long tensor
+    # ensure every gt matches with its prior of max overlap
+    for j in range(best_prior_idx.size(0)):  # 判别此anchor是预测哪一个boxes
+        best_truth_idx[best_prior_idx[j]] = j
+    matches = truths[best_truth_idx]  # Shape: [num_priors,4] 此处为每一个anchor对应的bbox取出来
+    conf = labels[best_truth_idx]  # Shape: [num_priors]      此处为每一个anchor对应的label取出来
+    conf[best_truth_overlap < threshold] = 0  # label as background   overlap<0.35的全部作为负样本
+    loc = encode(matches, priors, variances)
+
+    matches_landm = landms[best_truth_idx]
+    landm = encode_landm(matches_landm, priors, variances)
+    loc_t[idx] = loc  # [num_priors,4] encoded offsets to learn
+    conf_t[idx] = conf  # [num_priors] top class label for each prior
+    landm_t[idx] = landm
+
+
+def encode(matched, priors, variances):
+    """Encode the variances from the priorbox layers into the ground truth boxes
+    we have matched (based on jaccard overlap) with the prior boxes.
+    Args:
+        matched: (tensor) Coords of ground truth for each prior in point-form
+            Shape: [num_priors, 4].
+        priors: (tensor) Prior boxes in center-offset form
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        encoded boxes (tensor), Shape: [num_priors, 4]
+    """
+
+    # dist b/t match center and prior's center
+    g_cxcy = (matched[:, :2] + matched[:, 2:]) / 2 - priors[:, :2]
+    # encode variance
+    g_cxcy /= (variances[0] * priors[:, 2:])
+    # match wh / prior wh
+    g_wh = (matched[:, 2:] - matched[:, :2]) / priors[:, 2:]
+    g_wh = torch.log(g_wh) / variances[1]
+    # return target for smooth_l1_loss
+    return torch.cat([g_cxcy, g_wh], 1)  # [num_priors,4]
+
+
+def encode_landm(matched, priors, variances):
+    """Encode the variances from the priorbox layers into the ground truth boxes
+    we have matched (based on jaccard overlap) with the prior boxes.
+    Args:
+        matched: (tensor) Coords of ground truth for each prior in point-form
+            Shape: [num_priors, 10].
+        priors: (tensor) Prior boxes in center-offset form
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        encoded landm (tensor), Shape: [num_priors, 10]
+    """
+
+    # dist b/t match center and prior's center
+    matched = torch.reshape(matched, (matched.size(0), 5, 2))
+    priors_cx = priors[:, 0].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors_cy = priors[:, 1].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors_w = priors[:, 2].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors_h = priors[:, 3].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors = torch.cat([priors_cx, priors_cy, priors_w, priors_h], dim=2)
+    g_cxcy = matched[:, :, :2] - priors[:, :, :2]
+    # encode variance
+    g_cxcy /= (variances[0] * priors[:, :, 2:])
+    # g_cxcy /= priors[:, :, 2:]
+    g_cxcy = g_cxcy.reshape(g_cxcy.size(0), -1)
+    # return target for smooth_l1_loss
+    return g_cxcy
+
+
+# Adapted from https://github.com/Hakuyume/chainer-ssd
+def decode(loc, priors, variances):
+    """Decode locations from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        loc (tensor): location predictions for loc layers,
+            Shape: [num_priors,4]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded bounding box predictions
+    """
+
+    boxes = torch.cat((priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:],
+                       priors[:, 2:] * torch.exp(loc[:, 2:] * variances[1])), 1)
+    boxes[:, :2] -= boxes[:, 2:] / 2
+    boxes[:, 2:] += boxes[:, :2]
+    return boxes
+
+
+def decode_landm(pre, priors, variances):
+    """Decode landm from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        pre (tensor): landm predictions for loc layers,
+            Shape: [num_priors,10]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded landm predictions
+    """
+    tmp = (
+        priors[:, :2] + pre[:, :2] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 2:4] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 4:6] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 6:8] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 8:10] * variances[0] * priors[:, 2:],
+    )
+    landms = torch.cat(tmp, dim=1)
+    return landms
+
+
+def batched_decode(b_loc, priors, variances):
+    """Decode locations from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        b_loc (tensor): location predictions for loc layers,
+            Shape: [num_batches,num_priors,4]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [1,num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded bounding box predictions
+    """
+    boxes = (
+        priors[:, :, :2] + b_loc[:, :, :2] * variances[0] * priors[:, :, 2:],
+        priors[:, :, 2:] * torch.exp(b_loc[:, :, 2:] * variances[1]),
+    )
+    boxes = torch.cat(boxes, dim=2)
+
+    boxes[:, :, :2] -= boxes[:, :, 2:] / 2
+    boxes[:, :, 2:] += boxes[:, :, :2]
+    return boxes
+
+
+def batched_decode_landm(pre, priors, variances):
+    """Decode landm from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        pre (tensor): landm predictions for loc layers,
+            Shape: [num_batches,num_priors,10]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [1,num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded landm predictions
+    """
+    landms = (
+        priors[:, :, :2] + pre[:, :, :2] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 2:4] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 4:6] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 6:8] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 8:10] * variances[0] * priors[:, :, 2:],
+    )
+    landms = torch.cat(landms, dim=2)
+    return landms
+
+
+def log_sum_exp(x):
+    """Utility function for computing log_sum_exp while determining
+    This will be used to determine unaveraged confidence loss across
+    all examples in a batch.
+    Args:
+        x (Variable(tensor)): conf_preds from conf layers
+    """
+    x_max = x.data.max()
+    return torch.log(torch.sum(torch.exp(x - x_max), 1, keepdim=True)) + x_max
+
+
+# Original author: Francisco Massa:
+# https://github.com/fmassa/object-detection.torch
+# Ported to PyTorch by Max deGroot (02/01/2017)
+def nms(boxes, scores, overlap=0.5, top_k=200):
+    """Apply non-maximum suppression at test time to avoid detecting too many
+    overlapping bounding boxes for a given object.
+    Args:
+        boxes: (tensor) The location preds for the img, Shape: [num_priors,4].
+        scores: (tensor) The class predscores for the img, Shape:[num_priors].
+        overlap: (float) The overlap thresh for suppressing unnecessary boxes.
+        top_k: (int) The Maximum number of box preds to consider.
+    Return:
+        The indices of the kept boxes with respect to num_priors.
+    """
+
+    keep = torch.Tensor(scores.size(0)).fill_(0).long()
+    if boxes.numel() == 0:
+        return keep
+    x1 = boxes[:, 0]
+    y1 = boxes[:, 1]
+    x2 = boxes[:, 2]
+    y2 = boxes[:, 3]
+    area = torch.mul(x2 - x1, y2 - y1)
+    v, idx = scores.sort(0)  # sort in ascending order
+    # I = I[v >= 0.01]
+    idx = idx[-top_k:]  # indices of the top-k largest vals
+    xx1 = boxes.new()
+    yy1 = boxes.new()
+    xx2 = boxes.new()
+    yy2 = boxes.new()
+    w = boxes.new()
+    h = boxes.new()
+
+    # keep = torch.Tensor()
+    count = 0
+    while idx.numel() > 0:
+        i = idx[-1]  # index of current largest val
+        # keep.append(i)
+        keep[count] = i
+        count += 1
+        if idx.size(0) == 1:
+            break
+        idx = idx[:-1]  # remove kept element from view
+        # load bboxes of next highest vals
+        torch.index_select(x1, 0, idx, out=xx1)
+        torch.index_select(y1, 0, idx, out=yy1)
+        torch.index_select(x2, 0, idx, out=xx2)
+        torch.index_select(y2, 0, idx, out=yy2)
+        # store element-wise max with next highest score
+        xx1 = torch.clamp(xx1, min=x1[i])
+        yy1 = torch.clamp(yy1, min=y1[i])
+        xx2 = torch.clamp(xx2, max=x2[i])
+        yy2 = torch.clamp(yy2, max=y2[i])
+        w.resize_as_(xx2)
+        h.resize_as_(yy2)
+        w = xx2 - xx1
+        h = yy2 - yy1
+        # check sizes of xx1 and xx2.. after each iteration
+        w = torch.clamp(w, min=0.0)
+        h = torch.clamp(h, min=0.0)
+        inter = w * h
+        # IoU = i / (area(a) + area(b) - i)
+        rem_areas = torch.index_select(area, 0, idx)  # load remaining areas)
+        union = (rem_areas - inter) + area[i]
+        IoU = inter / union  # store result in iou
+        # keep only elements with an IoU <= overlap
+        idx = idx[IoU.le(overlap)]
+    return keep, count

facelib/detection/yolov5face/face_detector.py

@@ -0,0 +1,142 @@
+import copy
+import os
+from pathlib import Path
+
+import cv2
+import numpy as np
+import torch
+from torch import nn
+
+from facelib.detection.yolov5face.models.common import Conv
+from facelib.detection.yolov5face.models.yolo import Model
+from facelib.detection.yolov5face.utils.datasets import letterbox
+from facelib.detection.yolov5face.utils.general import (
+    check_img_size,
+    non_max_suppression_face,
+    scale_coords,
+    scale_coords_landmarks,
+)
+
+IS_HIGH_VERSION = tuple(map(int, torch.__version__.split('+')[0].split('.'))) >= (1, 9, 0)
+
+
+def isListempty(inList):
+    if isinstance(inList, list): # Is a list
+        return all(map(isListempty, inList))
+    return False # Not a list
+
+class YoloDetector:
+    def __init__(
+        self,
+        config_name,
+        min_face=10,
+        target_size=None,
+        device='cuda',
+    ):
+        """
+        config_name: name of .yaml config with network configuration from models/ folder.
+        min_face : minimal face size in pixels.
+        target_size : target size of smaller image axis (choose lower for faster work). e.g. 480, 720, 1080.
+                    None for original resolution.
+        """
+        self._class_path = Path(__file__).parent.absolute()
+        self.target_size = target_size
+        self.min_face = min_face
+        self.detector = Model(cfg=config_name)
+        self.device = device
+
+
+    def _preprocess(self, imgs):
+        """
+        Preprocessing image before passing through the network. Resize and conversion to torch tensor.
+        """
+        pp_imgs = []
+        for img in imgs:
+            h0, w0 = img.shape[:2]  # orig hw
+            if self.target_size:
+                r = self.target_size / min(h0, w0)  # resize image to img_size
+                if r < 1:
+                    img = cv2.resize(img, (int(w0 * r), int(h0 * r)), interpolation=cv2.INTER_LINEAR)
+
+            imgsz = check_img_size(max(img.shape[:2]), s=self.detector.stride.max())  # check img_size
+            img = letterbox(img, new_shape=imgsz)[0]
+            pp_imgs.append(img)
+        pp_imgs = np.array(pp_imgs)
+        pp_imgs = pp_imgs.transpose(0, 3, 1, 2)
+        pp_imgs = torch.from_numpy(pp_imgs).to(self.device)
+        pp_imgs = pp_imgs.float()  # uint8 to fp16/32
+        return pp_imgs / 255.0  # 0 - 255 to 0.0 - 1.0
+
+    def _postprocess(self, imgs, origimgs, pred, conf_thres, iou_thres):
+        """
+        Postprocessing of raw pytorch model output.
+        Returns:
+            bboxes: list of arrays with 4 coordinates of bounding boxes with format x1,y1,x2,y2.
+            points: list of arrays with coordinates of 5 facial keypoints (eyes, nose, lips corners).
+        """
+        bboxes = [[] for _ in range(len(origimgs))]
+        landmarks = [[] for _ in range(len(origimgs))]
+
+        pred = non_max_suppression_face(pred, conf_thres, iou_thres)
+
+        for image_id, origimg in enumerate(origimgs):
+            img_shape = origimg.shape
+            image_height, image_width = img_shape[:2]
+            gn = torch.tensor(img_shape)[[1, 0, 1, 0]]  # normalization gain whwh
+            gn_lks = torch.tensor(img_shape)[[1, 0, 1, 0, 1, 0, 1, 0, 1, 0]]  # normalization gain landmarks
+            det = pred[image_id].cpu()
+            scale_coords(imgs[image_id].shape[1:], det[:, :4], img_shape).round()
+            scale_coords_landmarks(imgs[image_id].shape[1:], det[:, 5:15], img_shape).round()
+
+            for j in range(det.size()[0]):
+                box = (det[j, :4].view(1, 4) / gn).view(-1).tolist()
+                box = list(
+                    map(int, [box[0] * image_width, box[1] * image_height, box[2] * image_width, box[3] * image_height])
+                )
+                if box[3] - box[1] < self.min_face:
+                    continue
+                lm = (det[j, 5:15].view(1, 10) / gn_lks).view(-1).tolist()
+                lm = list(map(int, [i * image_width if j % 2 == 0 else i * image_height for j, i in enumerate(lm)]))
+                lm = [lm[i : i + 2] for i in range(0, len(lm), 2)]
+                bboxes[image_id].append(box)
+                landmarks[image_id].append(lm)
+        return bboxes, landmarks
+
+    def detect_faces(self, imgs, conf_thres=0.7, iou_thres=0.5):
+        """
+        Get bbox coordinates and keypoints of faces on original image.
+        Params:
+            imgs: image or list of images to detect faces on with BGR order (convert to RGB order for inference)
+            conf_thres: confidence threshold for each prediction
+            iou_thres: threshold for NMS (filter of intersecting bboxes)
+        Returns:
+            bboxes: list of arrays with 4 coordinates of bounding boxes with format x1,y1,x2,y2.
+            points: list of arrays with coordinates of 5 facial keypoints (eyes, nose, lips corners).
+        """
+        # Pass input images through face detector
+        images = imgs if isinstance(imgs, list) else [imgs]
+        images = [cv2.cvtColor(img, cv2.COLOR_BGR2RGB) for img in images]
+        origimgs = copy.deepcopy(images)
+
+        images = self._preprocess(images)
+        
+        if IS_HIGH_VERSION:
+            with torch.inference_mode():  # for pytorch>=1.9 
+                pred = self.detector(images)[0]
+        else:
+            with torch.no_grad():  # for pytorch<1.9
+                pred = self.detector(images)[0]
+
+        bboxes, points = self._postprocess(images, origimgs, pred, conf_thres, iou_thres)
+
+        # return bboxes, points
+        if not isListempty(points):
+            bboxes = np.array(bboxes).reshape(-1,4)
+            points = np.array(points).reshape(-1,10)
+            padding = bboxes[:,0].reshape(-1,1)
+            return np.concatenate((bboxes, padding, points), axis=1)
+        else:
+            return None
+
+    def __call__(self, *args):
+        return self.predict(*args)

facelib/detection/yolov5face/models/common.py

@@ -0,0 +1,299 @@
+# This file contains modules common to various models
+
+import math
+
+import numpy as np
+import torch
+from torch import nn
+
+from facelib.detection.yolov5face.utils.datasets import letterbox
+from facelib.detection.yolov5face.utils.general import (
+    make_divisible,
+    non_max_suppression,
+    scale_coords,
+    xyxy2xywh,
+)
+
+
+def autopad(k, p=None):  # kernel, padding
+    # Pad to 'same'
+    if p is None:
+        p = k // 2 if isinstance(k, int) else [x // 2 for x in k]  # auto-pad
+    return p
+
+
+def channel_shuffle(x, groups):
+    batchsize, num_channels, height, width = x.data.size()
+    channels_per_group = torch.div(num_channels, groups, rounding_mode="trunc")
+
+    # reshape
+    x = x.view(batchsize, groups, channels_per_group, height, width)
+    x = torch.transpose(x, 1, 2).contiguous()
+
+    # flatten
+    return x.view(batchsize, -1, height, width)
+
+
+def DWConv(c1, c2, k=1, s=1, act=True):
+    # Depthwise convolution
+    return Conv(c1, c2, k, s, g=math.gcd(c1, c2), act=act)
+
+
+class Conv(nn.Module):
+    # Standard convolution
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
+        super().__init__()
+        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = nn.SiLU() if act is True else (act if isinstance(act, nn.Module) else nn.Identity())
+
+    def forward(self, x):
+        return self.act(self.bn(self.conv(x)))
+
+    def fuseforward(self, x):
+        return self.act(self.conv(x))
+
+
+class StemBlock(nn.Module):
+    def __init__(self, c1, c2, k=3, s=2, p=None, g=1, act=True):
+        super().__init__()
+        self.stem_1 = Conv(c1, c2, k, s, p, g, act)
+        self.stem_2a = Conv(c2, c2 // 2, 1, 1, 0)
+        self.stem_2b = Conv(c2 // 2, c2, 3, 2, 1)
+        self.stem_2p = nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True)
+        self.stem_3 = Conv(c2 * 2, c2, 1, 1, 0)
+
+    def forward(self, x):
+        stem_1_out = self.stem_1(x)
+        stem_2a_out = self.stem_2a(stem_1_out)
+        stem_2b_out = self.stem_2b(stem_2a_out)
+        stem_2p_out = self.stem_2p(stem_1_out)
+        return self.stem_3(torch.cat((stem_2b_out, stem_2p_out), 1))
+
+
+class Bottleneck(nn.Module):
+    # Standard bottleneck
+    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_, c2, 3, 1, g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class BottleneckCSP(nn.Module):
+    # CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
+        self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
+        self.cv4 = Conv(2 * c_, c2, 1, 1)
+        self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
+        self.act = nn.LeakyReLU(0.1, inplace=True)
+        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
+
+    def forward(self, x):
+        y1 = self.cv3(self.m(self.cv1(x)))
+        y2 = self.cv2(x)
+        return self.cv4(self.act(self.bn(torch.cat((y1, y2), dim=1))))
+
+
+class C3(nn.Module):
+    # CSP Bottleneck with 3 convolutions
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c1, c_, 1, 1)
+        self.cv3 = Conv(2 * c_, c2, 1)  # act=FReLU(c2)
+        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
+
+    def forward(self, x):
+        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), dim=1))
+
+
+class ShuffleV2Block(nn.Module):
+    def __init__(self, inp, oup, stride):
+        super().__init__()
+
+        if not 1 <= stride <= 3:
+            raise ValueError("illegal stride value")
+        self.stride = stride
+
+        branch_features = oup // 2
+
+        if self.stride > 1:
+            self.branch1 = nn.Sequential(
+                self.depthwise_conv(inp, inp, kernel_size=3, stride=self.stride, padding=1),
+                nn.BatchNorm2d(inp),
+                nn.Conv2d(inp, branch_features, kernel_size=1, stride=1, padding=0, bias=False),
+                nn.BatchNorm2d(branch_features),
+                nn.SiLU(),
+            )
+        else:
+            self.branch1 = nn.Sequential()
+
+        self.branch2 = nn.Sequential(
+            nn.Conv2d(
+                inp if (self.stride > 1) else branch_features,
+                branch_features,
+                kernel_size=1,
+                stride=1,
+                padding=0,
+                bias=False,
+            ),
+            nn.BatchNorm2d(branch_features),
+            nn.SiLU(),
+            self.depthwise_conv(branch_features, branch_features, kernel_size=3, stride=self.stride, padding=1),
+            nn.BatchNorm2d(branch_features),
+            nn.Conv2d(branch_features, branch_features, kernel_size=1, stride=1, padding=0, bias=False),
+            nn.BatchNorm2d(branch_features),
+            nn.SiLU(),
+        )
+
+    @staticmethod
+    def depthwise_conv(i, o, kernel_size, stride=1, padding=0, bias=False):
+        return nn.Conv2d(i, o, kernel_size, stride, padding, bias=bias, groups=i)
+
+    def forward(self, x):
+        if self.stride == 1:
+            x1, x2 = x.chunk(2, dim=1)
+            out = torch.cat((x1, self.branch2(x2)), dim=1)
+        else:
+            out = torch.cat((self.branch1(x), self.branch2(x)), dim=1)
+        out = channel_shuffle(out, 2)
+        return out
+
+
+class SPP(nn.Module):
+    # Spatial pyramid pooling layer used in YOLOv3-SPP
+    def __init__(self, c1, c2, k=(5, 9, 13)):
+        super().__init__()
+        c_ = c1 // 2  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
+        self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])
+
+    def forward(self, x):
+        x = self.cv1(x)
+        return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))
+
+
+class Focus(nn.Module):
+    # Focus wh information into c-space
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
+        super().__init__()
+        self.conv = Conv(c1 * 4, c2, k, s, p, g, act)
+
+    def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
+        return self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1))
+
+
+class Concat(nn.Module):
+    # Concatenate a list of tensors along dimension
+    def __init__(self, dimension=1):
+        super().__init__()
+        self.d = dimension
+
+    def forward(self, x):
+        return torch.cat(x, self.d)
+
+
+class NMS(nn.Module):
+    # Non-Maximum Suppression (NMS) module
+    conf = 0.25  # confidence threshold
+    iou = 0.45  # IoU threshold
+    classes = None  # (optional list) filter by class
+
+    def forward(self, x):
+        return non_max_suppression(x[0], conf_thres=self.conf, iou_thres=self.iou, classes=self.classes)
+
+
+class AutoShape(nn.Module):
+    # input-robust model wrapper for passing cv2/np/PIL/torch inputs. Includes preprocessing, inference and NMS
+    img_size = 640  # inference size (pixels)
+    conf = 0.25  # NMS confidence threshold
+    iou = 0.45  # NMS IoU threshold
+    classes = None  # (optional list) filter by class
+
+    def __init__(self, model):
+        super().__init__()
+        self.model = model.eval()
+
+    def autoshape(self):
+        print("autoShape already enabled, skipping... ")  # model already converted to model.autoshape()
+        return self
+
+    def forward(self, imgs, size=640, augment=False, profile=False):
+        # Inference from various sources. For height=720, width=1280, RGB images example inputs are:
+        #   OpenCV:          = cv2.imread('image.jpg')[:,:,::-1]  # HWC BGR to RGB x(720,1280,3)
+        #   PIL:             = Image.open('image.jpg')  # HWC x(720,1280,3)
+        #   numpy:           = np.zeros((720,1280,3))  # HWC
+        #   torch:           = torch.zeros(16,3,720,1280)  # BCHW
+        #   multiple:        = [Image.open('image1.jpg'), Image.open('image2.jpg'), ...]  # list of images
+
+        p = next(self.model.parameters())  # for device and type
+        if isinstance(imgs, torch.Tensor):  # torch
+            return self.model(imgs.to(p.device).type_as(p), augment, profile)  # inference
+
+        # Pre-process
+        n, imgs = (len(imgs), imgs) if isinstance(imgs, list) else (1, [imgs])  # number of images, list of images
+        shape0, shape1 = [], []  # image and inference shapes
+        for i, im in enumerate(imgs):
+            im = np.array(im)  # to numpy
+            if im.shape[0] < 5:  # image in CHW
+                im = im.transpose((1, 2, 0))  # reverse dataloader .transpose(2, 0, 1)
+            im = im[:, :, :3] if im.ndim == 3 else np.tile(im[:, :, None], 3)  # enforce 3ch input
+            s = im.shape[:2]  # HWC
+            shape0.append(s)  # image shape
+            g = size / max(s)  # gain
+            shape1.append([y * g for y in s])
+            imgs[i] = im  # update
+        shape1 = [make_divisible(x, int(self.stride.max())) for x in np.stack(shape1, 0).max(0)]  # inference shape
+        x = [letterbox(im, new_shape=shape1, auto=False)[0] for im in imgs]  # pad
+        x = np.stack(x, 0) if n > 1 else x[0][None]  # stack
+        x = np.ascontiguousarray(x.transpose((0, 3, 1, 2)))  # BHWC to BCHW
+        x = torch.from_numpy(x).to(p.device).type_as(p) / 255.0  # uint8 to fp16/32
+
+        # Inference
+        with torch.no_grad():
+            y = self.model(x, augment, profile)[0]  # forward
+        y = non_max_suppression(y, conf_thres=self.conf, iou_thres=self.iou, classes=self.classes)  # NMS
+
+        # Post-process
+        for i in range(n):
+            scale_coords(shape1, y[i][:, :4], shape0[i])
+
+        return Detections(imgs, y, self.names)
+
+
+class Detections:
+    # detections class for YOLOv5 inference results
+    def __init__(self, imgs, pred, names=None):
+        super().__init__()
+        d = pred[0].device  # device
+        gn = [torch.tensor([*(im.shape[i] for i in [1, 0, 1, 0]), 1.0, 1.0], device=d) for im in imgs]  # normalizations
+        self.imgs = imgs  # list of images as numpy arrays
+        self.pred = pred  # list of tensors pred[0] = (xyxy, conf, cls)
+        self.names = names  # class names
+        self.xyxy = pred  # xyxy pixels
+        self.xywh = [xyxy2xywh(x) for x in pred]  # xywh pixels
+        self.xyxyn = [x / g for x, g in zip(self.xyxy, gn)]  # xyxy normalized
+        self.xywhn = [x / g for x, g in zip(self.xywh, gn)]  # xywh normalized
+        self.n = len(self.pred)
+
+    def __len__(self):
+        return self.n
+
+    def tolist(self):
+        # return a list of Detections objects, i.e. 'for result in results.tolist():'
+        x = [Detections([self.imgs[i]], [self.pred[i]], self.names) for i in range(self.n)]
+        for d in x:
+            for k in ["imgs", "pred", "xyxy", "xyxyn", "xywh", "xywhn"]:
+                setattr(d, k, getattr(d, k)[0])  # pop out of list
+        return x

facelib/detection/yolov5face/models/experimental.py

@@ -0,0 +1,45 @@
+# # This file contains experimental modules
+
+import numpy as np
+import torch
+from torch import nn
+
+from facelib.detection.yolov5face.models.common import Conv
+
+
+class CrossConv(nn.Module):
+    # Cross Convolution Downsample
+    def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False):
+        # ch_in, ch_out, kernel, stride, groups, expansion, shortcut
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, (1, k), (1, s))
+        self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class MixConv2d(nn.Module):
+    # Mixed Depthwise Conv https://arxiv.org/abs/1907.09595
+    def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True):
+        super().__init__()
+        groups = len(k)
+        if equal_ch:  # equal c_ per group
+            i = torch.linspace(0, groups - 1e-6, c2).floor()  # c2 indices
+            c_ = [(i == g).sum() for g in range(groups)]  # intermediate channels
+        else:  # equal weight.numel() per group
+            b = [c2] + [0] * groups
+            a = np.eye(groups + 1, groups, k=-1)
+            a -= np.roll(a, 1, axis=1)
+            a *= np.array(k) ** 2
+            a[0] = 1
+            c_ = np.linalg.lstsq(a, b, rcond=None)[0].round()  # solve for equal weight indices, ax = b
+
+        self.m = nn.ModuleList([nn.Conv2d(c1, int(c_[g]), k[g], s, k[g] // 2, bias=False) for g in range(groups)])
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = nn.LeakyReLU(0.1, inplace=True)
+
+    def forward(self, x):
+        return x + self.act(self.bn(torch.cat([m(x) for m in self.m], 1)))

facelib/detection/yolov5face/models/yolo.py

@@ -0,0 +1,235 @@
+import math
+from copy import deepcopy
+from pathlib import Path
+
+import torch
+import yaml  # for torch hub
+from torch import nn
+
+from facelib.detection.yolov5face.models.common import (
+    C3,
+    NMS,
+    SPP,
+    AutoShape,
+    Bottleneck,
+    BottleneckCSP,
+    Concat,
+    Conv,
+    DWConv,
+    Focus,
+    ShuffleV2Block,
+    StemBlock,
+)
+from facelib.detection.yolov5face.models.experimental import CrossConv, MixConv2d
+from facelib.detection.yolov5face.utils.autoanchor import check_anchor_order
+from facelib.detection.yolov5face.utils.general import make_divisible
+from facelib.detection.yolov5face.utils.torch_utils import copy_attr, fuse_conv_and_bn
+
+
+class Detect(nn.Module):
+    stride = None  # strides computed during build
+    export = False  # onnx export
+
+    def __init__(self, nc=80, anchors=(), ch=()):  # detection layer
+        super().__init__()
+        self.nc = nc  # number of classes
+        self.no = nc + 5 + 10  # number of outputs per anchor
+
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer("anchors", a)  # shape(nl,na,2)
+        self.register_buffer("anchor_grid", a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+
+    def forward(self, x):
+        z = []  # inference output
+        if self.export:
+            for i in range(self.nl):
+                x[i] = self.m[i](x[i])
+            return x
+        for i in range(self.nl):
+            x[i] = self.m[i](x[i])  # conv
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = torch.full_like(x[i], 0)
+                y[..., [0, 1, 2, 3, 4, 15]] = x[i][..., [0, 1, 2, 3, 4, 15]].sigmoid()
+                y[..., 5:15] = x[i][..., 5:15]
+
+                y[..., 0:2] = (y[..., 0:2] * 2.0 - 0.5 + self.grid[i].to(x[i].device)) * self.stride[i]  # xy
+                y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+
+                y[..., 5:7] = (
+                    y[..., 5:7] * self.anchor_grid[i] + self.grid[i].to(x[i].device) * self.stride[i]
+                )  # landmark x1 y1
+                y[..., 7:9] = (
+                    y[..., 7:9] * self.anchor_grid[i] + self.grid[i].to(x[i].device) * self.stride[i]
+                )  # landmark x2 y2
+                y[..., 9:11] = (
+                    y[..., 9:11] * self.anchor_grid[i] + self.grid[i].to(x[i].device) * self.stride[i]
+                )  # landmark x3 y3
+                y[..., 11:13] = (
+                    y[..., 11:13] * self.anchor_grid[i] + self.grid[i].to(x[i].device) * self.stride[i]
+                )  # landmark x4 y4
+                y[..., 13:15] = (
+                    y[..., 13:15] * self.anchor_grid[i] + self.grid[i].to(x[i].device) * self.stride[i]
+                )  # landmark x5 y5
+
+                z.append(y.view(bs, -1, self.no))
+
+        return x if self.training else (torch.cat(z, 1), x)
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        # yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)], indexing="ij") # for pytorch>=1.10
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+
+class Model(nn.Module):
+    def __init__(self, cfg="yolov5s.yaml", ch=3, nc=None):  # model, input channels, number of classes
+        super().__init__()
+        self.yaml_file = Path(cfg).name
+        with Path(cfg).open(encoding="utf8") as f:
+            self.yaml = yaml.safe_load(f)  # model dict
+
+        # Define model
+        ch = self.yaml["ch"] = self.yaml.get("ch", ch)  # input channels
+        if nc and nc != self.yaml["nc"]:
+            self.yaml["nc"] = nc  # override yaml value
+
+        self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch])  # model, savelist
+        self.names = [str(i) for i in range(self.yaml["nc"])]  # default names
+
+        # Build strides, anchors
+        m = self.model[-1]  # Detect()
+        if isinstance(m, Detect):
+            s = 128  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
+            m.anchors /= m.stride.view(-1, 1, 1)
+            check_anchor_order(m)
+            self.stride = m.stride
+            self._initialize_biases()  # only run once
+
+    def forward(self, x):
+        return self.forward_once(x)  # single-scale inference, train
+
+    def forward_once(self, x):
+        y = []  # outputs
+        for m in self.model:
+            if m.f != -1:  # if not from previous layer
+                x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]  # from earlier layers
+
+            x = m(x)  # run
+            y.append(x if m.i in self.save else None)  # save output
+
+        return x
+
+    def _initialize_biases(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # https://arxiv.org/abs/1708.02002 section 3.3
+        m = self.model[-1]  # Detect() module
+        for mi, s in zip(m.m, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b.data[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+
+    def _print_biases(self):
+        m = self.model[-1]  # Detect() module
+        for mi in m.m:  # from
+            b = mi.bias.detach().view(m.na, -1).T  # conv.bias(255) to (3,85)
+            print(("%6g Conv2d.bias:" + "%10.3g" * 6) % (mi.weight.shape[1], *b[:5].mean(1).tolist(), b[5:].mean()))
+
+    def fuse(self):  # fuse model Conv2d() + BatchNorm2d() layers
+        print("Fusing layers... ")
+        for m in self.model.modules():
+            if isinstance(m, Conv) and hasattr(m, "bn"):
+                m.conv = fuse_conv_and_bn(m.conv, m.bn)  # update conv
+                delattr(m, "bn")  # remove batchnorm
+                m.forward = m.fuseforward  # update forward
+            elif type(m) is nn.Upsample:
+                m.recompute_scale_factor = None  # torch 1.11.0 compatibility
+        return self
+
+    def nms(self, mode=True):  # add or remove NMS module
+        present = isinstance(self.model[-1], NMS)  # last layer is NMS
+        if mode and not present:
+            print("Adding NMS... ")
+            m = NMS()  # module
+            m.f = -1  # from
+            m.i = self.model[-1].i + 1  # index
+            self.model.add_module(name=str(m.i), module=m)  # add
+            self.eval()
+        elif not mode and present:
+            print("Removing NMS... ")
+            self.model = self.model[:-1]  # remove
+        return self
+
+    def autoshape(self):  # add autoShape module
+        print("Adding autoShape... ")
+        m = AutoShape(self)  # wrap model
+        copy_attr(m, self, include=("yaml", "nc", "hyp", "names", "stride"), exclude=())  # copy attributes
+        return m
+
+
+def parse_model(d, ch):  # model_dict, input_channels(3)
+    anchors, nc, gd, gw = d["anchors"], d["nc"], d["depth_multiple"], d["width_multiple"]
+    na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors
+    no = na * (nc + 5)  # number of outputs = anchors * (classes + 5)
+
+    layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
+    for i, (f, n, m, args) in enumerate(d["backbone"] + d["head"]):  # from, number, module, args
+        m = eval(m) if isinstance(m, str) else m  # eval strings
+        for j, a in enumerate(args):
+            try:
+                args[j] = eval(a) if isinstance(a, str) else a  # eval strings
+            except:
+                pass
+
+        n = max(round(n * gd), 1) if n > 1 else n  # depth gain
+        if m in [
+            Conv,
+            Bottleneck,
+            SPP,
+            DWConv,
+            MixConv2d,
+            Focus,
+            CrossConv,
+            BottleneckCSP,
+            C3,
+            ShuffleV2Block,
+            StemBlock,
+        ]:
+            c1, c2 = ch[f], args[0]
+
+            c2 = make_divisible(c2 * gw, 8) if c2 != no else c2
+
+            args = [c1, c2, *args[1:]]
+            if m in [BottleneckCSP, C3]:
+                args.insert(2, n)
+                n = 1
+        elif m is nn.BatchNorm2d:
+            args = [ch[f]]
+        elif m is Concat:
+            c2 = sum(ch[-1 if x == -1 else x + 1] for x in f)
+        elif m is Detect:
+            args.append([ch[x + 1] for x in f])
+            if isinstance(args[1], int):  # number of anchors
+                args[1] = [list(range(args[1] * 2))] * len(f)
+        else:
+            c2 = ch[f]
+
+        m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args)  # module
+        t = str(m)[8:-2].replace("__main__.", "")  # module type
+        np = sum(x.numel() for x in m_.parameters())  # number params
+        m_.i, m_.f, m_.type, m_.np = i, f, t, np  # attach index, 'from' index, type, number params
+        save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist
+        layers.append(m_)
+        ch.append(c2)
+    return nn.Sequential(*layers), sorted(save)

facelib/detection/yolov5face/models/yolov5l.yaml

@@ -0,0 +1,47 @@
+# parameters
+nc: 1  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [4,5,  8,10,  13,16]  # P3/8
+  - [23,29,  43,55,  73,105]  # P4/16
+  - [146,217,  231,300,  335,433]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, StemBlock, [64, 3, 2]],  # 0-P1/2
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],      # 2-P3/8
+   [-1, 9, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],      # 4-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],     # 6-P5/32
+   [-1, 1, SPP, [1024, [3,5,7]]],
+   [-1, 3, C3, [1024, False]],      # 8
+  ]
+
+# YOLOv5 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 5], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 12
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 3], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 16 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 13], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 19 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 9], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 22 (P5/32-large)
+
+   [[16, 19, 22], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

facelib/detection/yolov5face/models/yolov5n.yaml

@@ -0,0 +1,45 @@
+# parameters
+nc: 1  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [4,5,  8,10,  13,16]  # P3/8
+  - [23,29,  43,55,  73,105]  # P4/16
+  - [146,217,  231,300,  335,433]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, StemBlock, [32, 3, 2]],    # 0-P2/4
+   [-1, 1, ShuffleV2Block, [128, 2]], # 1-P3/8
+   [-1, 3, ShuffleV2Block, [128, 1]], # 2
+   [-1, 1, ShuffleV2Block, [256, 2]], # 3-P4/16
+   [-1, 7, ShuffleV2Block, [256, 1]], # 4
+   [-1, 1, ShuffleV2Block, [512, 2]], # 5-P5/32
+   [-1, 3, ShuffleV2Block, [512, 1]], # 6
+  ]
+
+# YOLOv5 head
+head:
+  [[-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, C3, [128, False]],  # 10
+
+   [-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 2], 1, Concat, [1]],  # cat backbone P3
+   [-1, 1, C3, [128, False]],  # 14 (P3/8-small)
+
+   [-1, 1, Conv, [128, 3, 2]],
+   [[-1, 11], 1, Concat, [1]],  # cat head P4
+   [-1, 1, C3, [128, False]],  # 17 (P4/16-medium)
+
+   [-1, 1, Conv, [128, 3, 2]],
+   [[-1, 7], 1, Concat, [1]],  # cat head P5
+   [-1, 1, C3, [128, False]],  # 20 (P5/32-large)
+
+   [[14, 17, 20], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

facelib/detection/yolov5face/utils/autoanchor.py

@@ -0,0 +1,12 @@
+# Auto-anchor utils
+
+
+def check_anchor_order(m):
+    # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary
+    a = m.anchor_grid.prod(-1).view(-1)  # anchor area
+    da = a[-1] - a[0]  # delta a
+    ds = m.stride[-1] - m.stride[0]  # delta s
+    if da.sign() != ds.sign():  # same order
+        print("Reversing anchor order")
+        m.anchors[:] = m.anchors.flip(0)
+        m.anchor_grid[:] = m.anchor_grid.flip(0)

facelib/detection/yolov5face/utils/datasets.py

@@ -0,0 +1,35 @@
+import cv2
+import numpy as np
+
+
+def letterbox(img, new_shape=(640, 640), color=(114, 114, 114), auto=True, scale_fill=False, scaleup=True):
+    # Resize image to a 32-pixel-multiple rectangle https://github.com/ultralytics/yolov3/issues/232
+    shape = img.shape[:2]  # current shape [height, width]
+    if isinstance(new_shape, int):
+        new_shape = (new_shape, new_shape)
+
+    # Scale ratio (new / old)
+    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
+    if not scaleup:  # only scale down, do not scale up (for better test mAP)
+        r = min(r, 1.0)
+
+    # Compute padding
+    ratio = r, r  # width, height ratios
+    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
+    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding
+    if auto:  # minimum rectangle
+        dw, dh = np.mod(dw, 64), np.mod(dh, 64)  # wh padding
+    elif scale_fill:  # stretch
+        dw, dh = 0.0, 0.0
+        new_unpad = (new_shape[1], new_shape[0])
+        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios
+
+    dw /= 2  # divide padding into 2 sides
+    dh /= 2
+
+    if shape[::-1] != new_unpad:  # resize
+        img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR)
+    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
+    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
+    img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
+    return img, ratio, (dw, dh)

facelib/detection/yolov5face/utils/extract_ckpt.py

@@ -0,0 +1,5 @@
+import torch
+import sys
+sys.path.insert(0,'./facelib/detection/yolov5face')
+model = torch.load('facelib/detection/yolov5face/yolov5n-face.pt', map_location='cpu')['model']
+torch.save(model.state_dict(),'weights/facelib/yolov5n-face.pth')

facelib/detection/yolov5face/utils/general.py

@@ -0,0 +1,271 @@
+import math
+import time
+
+import numpy as np
+import torch
+import torchvision
+
+
+def check_img_size(img_size, s=32):
+    # Verify img_size is a multiple of stride s
+    new_size = make_divisible(img_size, int(s))  # ceil gs-multiple
+    # if new_size != img_size:
+    #     print(f"WARNING: --img-size {img_size:g} must be multiple of max stride {s:g}, updating to {new_size:g}")
+    return new_size
+
+
+def make_divisible(x, divisor):
+    # Returns x evenly divisible by divisor
+    return math.ceil(x / divisor) * divisor
+
+
+def xyxy2xywh(x):
+    # Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] where xy1=top-left, xy2=bottom-right
+    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+    y[:, 0] = (x[:, 0] + x[:, 2]) / 2  # x center
+    y[:, 1] = (x[:, 1] + x[:, 3]) / 2  # y center
+    y[:, 2] = x[:, 2] - x[:, 0]  # width
+    y[:, 3] = x[:, 3] - x[:, 1]  # height
+    return y
+
+
+def xywh2xyxy(x):
+    # Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
+    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+    y[:, 0] = x[:, 0] - x[:, 2] / 2  # top left x
+    y[:, 1] = x[:, 1] - x[:, 3] / 2  # top left y
+    y[:, 2] = x[:, 0] + x[:, 2] / 2  # bottom right x
+    y[:, 3] = x[:, 1] + x[:, 3] / 2  # bottom right y
+    return y
+
+
+def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
+    # Rescale coords (xyxy) from img1_shape to img0_shape
+    if ratio_pad is None:  # calculate from img0_shape
+        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
+        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
+    else:
+        gain = ratio_pad[0][0]
+        pad = ratio_pad[1]
+
+    coords[:, [0, 2]] -= pad[0]  # x padding
+    coords[:, [1, 3]] -= pad[1]  # y padding
+    coords[:, :4] /= gain
+    clip_coords(coords, img0_shape)
+    return coords
+
+
+def clip_coords(boxes, img_shape):
+    # Clip bounding xyxy bounding boxes to image shape (height, width)
+    boxes[:, 0].clamp_(0, img_shape[1])  # x1
+    boxes[:, 1].clamp_(0, img_shape[0])  # y1
+    boxes[:, 2].clamp_(0, img_shape[1])  # x2
+    boxes[:, 3].clamp_(0, img_shape[0])  # y2
+
+
+def box_iou(box1, box2):
+    # https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
+    """
+    Return intersection-over-union (Jaccard index) of boxes.
+    Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
+    Arguments:
+        box1 (Tensor[N, 4])
+        box2 (Tensor[M, 4])
+    Returns:
+        iou (Tensor[N, M]): the NxM matrix containing the pairwise
+            IoU values for every element in boxes1 and boxes2
+    """
+
+    def box_area(box):
+        return (box[2] - box[0]) * (box[3] - box[1])
+
+    area1 = box_area(box1.T)
+    area2 = box_area(box2.T)
+
+    inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
+    return inter / (area1[:, None] + area2 - inter)
+
+
+def non_max_suppression_face(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, labels=()):
+    """Performs Non-Maximum Suppression (NMS) on inference results
+    Returns:
+         detections with shape: nx6 (x1, y1, x2, y2, conf, cls)
+    """
+
+    nc = prediction.shape[2] - 15  # number of classes
+    xc = prediction[..., 4] > conf_thres  # candidates
+
+    # Settings
+    # (pixels) maximum box width and height
+    max_wh = 4096
+    time_limit = 10.0  # seconds to quit after
+    redundant = True  # require redundant detections
+    multi_label = nc > 1  # multiple labels per box (adds 0.5ms/img)
+    merge = False  # use merge-NMS
+
+    t = time.time()
+    output = [torch.zeros((0, 16), device=prediction.device)] * prediction.shape[0]
+    for xi, x in enumerate(prediction):  # image index, image inference
+        # Apply constraints
+        x = x[xc[xi]]  # confidence
+
+        # Cat apriori labels if autolabelling
+        if labels and len(labels[xi]):
+            label = labels[xi]
+            v = torch.zeros((len(label), nc + 15), device=x.device)
+            v[:, :4] = label[:, 1:5]  # box
+            v[:, 4] = 1.0  # conf
+            v[range(len(label)), label[:, 0].long() + 15] = 1.0  # cls
+            x = torch.cat((x, v), 0)
+
+        # If none remain process next image
+        if not x.shape[0]:
+            continue
+
+        # Compute conf
+        x[:, 15:] *= x[:, 4:5]  # conf = obj_conf * cls_conf
+
+        # Box (center x, center y, width, height) to (x1, y1, x2, y2)
+        box = xywh2xyxy(x[:, :4])
+
+        # Detections matrix nx6 (xyxy, conf, landmarks, cls)
+        if multi_label:
+            i, j = (x[:, 15:] > conf_thres).nonzero(as_tuple=False).T
+            x = torch.cat((box[i], x[i, j + 15, None], x[:, 5:15], j[:, None].float()), 1)
+        else:  # best class only
+            conf, j = x[:, 15:].max(1, keepdim=True)
+            x = torch.cat((box, conf, x[:, 5:15], j.float()), 1)[conf.view(-1) > conf_thres]
+
+        # Filter by class
+        if classes is not None:
+            x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
+
+        # If none remain process next image
+        n = x.shape[0]  # number of boxes
+        if not n:
+            continue
+
+        # Batched NMS
+        c = x[:, 15:16] * (0 if agnostic else max_wh)  # classes
+        boxes, scores = x[:, :4] + c, x[:, 4]  # boxes (offset by class), scores
+        i = torchvision.ops.nms(boxes, scores, iou_thres)  # NMS
+
+        if merge and (1 < n < 3e3):  # Merge NMS (boxes merged using weighted mean)
+            # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
+            iou = box_iou(boxes[i], boxes) > iou_thres  # iou matrix
+            weights = iou * scores[None]  # box weights
+            x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True)  # merged boxes
+            if redundant:
+                i = i[iou.sum(1) > 1]  # require redundancy
+
+        output[xi] = x[i]
+        if (time.time() - t) > time_limit:
+            break  # time limit exceeded
+
+    return output
+
+
+def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, labels=()):
+    """Performs Non-Maximum Suppression (NMS) on inference results
+
+    Returns:
+         detections with shape: nx6 (x1, y1, x2, y2, conf, cls)
+    """
+
+    nc = prediction.shape[2] - 5  # number of classes
+    xc = prediction[..., 4] > conf_thres  # candidates
+
+    # Settings
+    # (pixels) maximum box width and height
+    max_wh = 4096
+    time_limit = 10.0  # seconds to quit after
+    redundant = True  # require redundant detections
+    multi_label = nc > 1  # multiple labels per box (adds 0.5ms/img)
+    merge = False  # use merge-NMS
+
+    t = time.time()
+    output = [torch.zeros((0, 6), device=prediction.device)] * prediction.shape[0]
+    for xi, x in enumerate(prediction):  # image index, image inference
+        x = x[xc[xi]]  # confidence
+
+        # Cat apriori labels if autolabelling
+        if labels and len(labels[xi]):
+            label_id = labels[xi]
+            v = torch.zeros((len(label_id), nc + 5), device=x.device)
+            v[:, :4] = label_id[:, 1:5]  # box
+            v[:, 4] = 1.0  # conf
+            v[range(len(label_id)), label_id[:, 0].long() + 5] = 1.0  # cls
+            x = torch.cat((x, v), 0)
+
+        # If none remain process next image
+        if not x.shape[0]:
+            continue
+
+        # Compute conf
+        x[:, 5:] *= x[:, 4:5]  # conf = obj_conf * cls_conf
+
+        # Box (center x, center y, width, height) to (x1, y1, x2, y2)
+        box = xywh2xyxy(x[:, :4])
+
+        # Detections matrix nx6 (xyxy, conf, cls)
+        if multi_label:
+            i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
+            x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
+        else:  # best class only
+            conf, j = x[:, 5:].max(1, keepdim=True)
+            x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
+
+        # Filter by class
+        if classes is not None:
+            x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
+
+        # Check shape
+        n = x.shape[0]  # number of boxes
+        if not n:  # no boxes
+            continue
+
+        x = x[x[:, 4].argsort(descending=True)]  # sort by confidence
+
+        # Batched NMS
+        c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes
+        boxes, scores = x[:, :4] + c, x[:, 4]  # boxes (offset by class), scores
+        i = torchvision.ops.nms(boxes, scores, iou_thres)  # NMS
+        if merge and (1 < n < 3e3):  # Merge NMS (boxes merged using weighted mean)
+            # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
+            iou = box_iou(boxes[i], boxes) > iou_thres  # iou matrix
+            weights = iou * scores[None]  # box weights
+            x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True)  # merged boxes
+            if redundant:
+                i = i[iou.sum(1) > 1]  # require redundancy
+
+        output[xi] = x[i]
+        if (time.time() - t) > time_limit:
+            print(f"WARNING: NMS time limit {time_limit}s exceeded")
+            break  # time limit exceeded
+
+    return output
+
+
+def scale_coords_landmarks(img1_shape, coords, img0_shape, ratio_pad=None):
+    # Rescale coords (xyxy) from img1_shape to img0_shape
+    if ratio_pad is None:  # calculate from img0_shape
+        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
+        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
+    else:
+        gain = ratio_pad[0][0]
+        pad = ratio_pad[1]
+
+    coords[:, [0, 2, 4, 6, 8]] -= pad[0]  # x padding
+    coords[:, [1, 3, 5, 7, 9]] -= pad[1]  # y padding
+    coords[:, :10] /= gain
+    coords[:, 0].clamp_(0, img0_shape[1])  # x1
+    coords[:, 1].clamp_(0, img0_shape[0])  # y1
+    coords[:, 2].clamp_(0, img0_shape[1])  # x2
+    coords[:, 3].clamp_(0, img0_shape[0])  # y2
+    coords[:, 4].clamp_(0, img0_shape[1])  # x3
+    coords[:, 5].clamp_(0, img0_shape[0])  # y3
+    coords[:, 6].clamp_(0, img0_shape[1])  # x4
+    coords[:, 7].clamp_(0, img0_shape[0])  # y4
+    coords[:, 8].clamp_(0, img0_shape[1])  # x5
+    coords[:, 9].clamp_(0, img0_shape[0])  # y5
+    return coords

facelib/detection/yolov5face/utils/torch_utils.py

@@ -0,0 +1,40 @@
+import torch
+from torch import nn
+
+
+def fuse_conv_and_bn(conv, bn):
+    # Fuse convolution and batchnorm layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/
+    fusedconv = (
+        nn.Conv2d(
+            conv.in_channels,
+            conv.out_channels,
+            kernel_size=conv.kernel_size,
+            stride=conv.stride,
+            padding=conv.padding,
+            groups=conv.groups,
+            bias=True,
+        )
+        .requires_grad_(False)
+        .to(conv.weight.device)
+    )
+
+    # prepare filters
+    w_conv = conv.weight.clone().view(conv.out_channels, -1)
+    w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
+    fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.size()))
+
+    # prepare spatial bias
+    b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias
+    b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
+    fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
+
+    return fusedconv
+
+
+def copy_attr(a, b, include=(), exclude=()):
+    # Copy attributes from b to a, options to only include [...] and to exclude [...]
+    for k, v in b.__dict__.items():
+        if (include and k not in include) or k.startswith("_") or k in exclude:
+            continue
+
+        setattr(a, k, v)