Fix GPU detection on Apple Silicon

README.md

@@ -64,14 +64,12 @@ NeoRefacer has been tested on the following operating systems:
 
 | Operating System | CPU Support | GPU Support |
 | ---------------- | ----------- | ----------- |
-| MacOSX           | ✅         | :warning:         |
+| MacOSX           | ✅         | ✅         |
 | Windows          | ✅         | ✅         |
 | Linux            | ✅         | ✅         |
 
 The application is compatible with both CPU and GPU (Nvidia CUDA) environments, and MacOSX (CoreML) 
 
-:warning: Please note, we do not recommend using `onnxruntime-silicon` on MacOSX due to an apparent issue with memory management. If you manage to compile `onnxruntime` for Silicon, the program is prepared to use CoreML.
-
 ## Installation
 
 NeoRefacer has been tested and is known to work with Python 3.11.11, but it is likely to work with other Python versions as well. It is recommended to use a virtual environment, such as [Conda](https://www.anaconda.com/download), for setting up and running the project to avoid potential conflicts with other Python packages you may have installed.
@@ -82,10 +80,13 @@ Follow these steps to install Refacer and its dependencies:
 
 ```bash
     # Check if ffmpeg is available (if not, you might to download it and add it to your PATH)
-    # Windows: download ffmpeg-git-essentials.7z from https://www.gyan.dev/ffmpeg/builds/
-    # Other systems: see a tutorial https://www.hostinger.com/tutorials/how-to-install-ffmpeg
     ffmpeg  
     
+    # Windows: download ffmpeg-git-essentials.7z from https://www.gyan.dev/ffmpeg/builds/
+    # MacOS: if you have brew installed:
+    # brew install ffmpeg
+    # Other systems: see a tutorial https://www.hostinger.com/tutorials/how-to-install-ffmpeg 
+
     # Clone the repository
     git clone https://github.com/MechasAI/NeoRefacer.git
     cd NeoRefacer

refacer.py

@@ -112,38 +112,30 @@ class Refacer:
             raise Exception("ERROR, something went wrong downloading the model!")
 
     def __check_providers(self):
+        available_providers = rt.get_available_providers()
+
         if self.force_cpu:
             self.providers = ['CPUExecutionProvider']
         else:
-            self.providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+            # Prefer faster execution providers in order
+            self.providers = []
+            for p in ['CoreMLExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']:
+                if p in available_providers:
+                    self.providers.append(p)
 
         rt.set_default_logger_severity(4)
         self.sess_options = rt.SessionOptions()
-        self.sess_options.execution_mode = rt.ExecutionMode.ORT_PARALLEL  # Better parallelism
+        self.sess_options.execution_mode = rt.ExecutionMode.ORT_PARALLEL
         self.sess_options.graph_optimization_level = rt.GraphOptimizationLevel.ORT_ENABLE_ALL
 
-        def __check_providers(self):
-            if self.force_cpu:
-                self.providers = ['CPUExecutionProvider']
-            else:
-                self.providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
-
-        rt.set_default_logger_severity(4)
-
-        self.sess_options = rt.SessionOptions()
-        self.sess_options.execution_mode = rt.ExecutionMode.ORT_PARALLEL  # Better parallelism
-        self.sess_options.graph_optimization_level = rt.GraphOptimizationLevel.ORT_ENABLE_ALL
-
-        # Use a temporary model session to detect the actual active provider
         test_model = os.path.expanduser("~/.insightface/models/buffalo_l/det_10g.onnx")
         try:
             test_session = rt.InferenceSession(test_model, self.sess_options, providers=self.providers)
-          active_provider = test_session.get_providers()[0]  # First provider used
+            active_provider = test_session.get_providers()[0]
         except Exception as e:
             print(f"[ERROR] Failed to create test session: {e}")
-           active_provider = 'CPUExecutionProvider'  # Safe fallback
+            active_provider = 'CPUExecutionProvider'
 
-         # Set mode based on actual provider
         if active_provider == 'CUDAExecutionProvider':
             self.mode = RefacerMode.CUDA
             self.use_num_cpus = 2
@@ -152,7 +144,6 @@ class Refacer:
             self.mode = RefacerMode.COREML
             self.use_num_cpus = max(mp.cpu_count() - 1, 1)
             self.sess_options.intra_op_num_threads = int(self.use_num_cpus / 2)
-        # elif active_provider == 'TensorrtExecutionProvider':
         elif self.colab_performance:
             self.mode = RefacerMode.TENSORRT
             self.use_num_cpus = max(mp.cpu_count() - 1, 1)
@@ -162,6 +153,7 @@ class Refacer:
             self.use_num_cpus = max(mp.cpu_count() - 1, 1)
             self.sess_options.intra_op_num_threads = int(self.use_num_cpus / 2)
 
+        print(f"Available providers: {available_providers}")
         print(f"Using providers: {self.providers}")
         print(f"Active provider: {active_provider}")
         print(f"Mode: {self.mode}")

requirements-COREML.txt

@@ -5,7 +5,7 @@ gradio==5.22.0
 insightface==0.7.3
 numpy==1.24.3
 onnx==1.14.0
-onnxruntime-silicon
+onnxruntime-silicon==1.16.3
 opencv_python==4.7.0.72
 opencv_python_headless==4.7.0.72
 scikit-image==0.20.0