Fork of Linzaer/Ultra-Light-Fast-Generic-Face-Detector-1MB

Differences between original repository and fork:

• Compatibility with PyTorch >=2.4. (🔥)
• Original pretrained models and converted ONNX models from GitHub releases page. (🔥)
• Sample script detect_img.py for inference of single image.
• Added command line arguments to converttoonnx.py.
• The following deprecations and errors has been fixed:
  • DeprecationWarning: 'np.float' is a deprecated alias for builtin 'float'.
  • FutureWarning: You are using 'torch.load' with 'weights_only=False'.
  • FutureWarning: Cython directive 'language_level' not set.
  • Cython Warning: Using deprecated NumPy API.
  • cv2.error: Can't parse 'xx'. Sequence item with index 0 has a wrong type.
  • cv2.error: Can't parse 'xx'. Expected sequence length 4, got 2.

Installation

shell pip install -r requirements.txt

Pretrained models

• Download links:

| Name | Model Size (MB) | Link | SHA-256 | |--------------------|-----------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------| | UltraFace-slim-320 | 1.0
1.1 | PyTorch
ONNX | cd24abce45da5dbc7cfd8167cd3d5f955382dfc9d9ae9459f0026abd3c2e38a4
3a85d818ae391c030694f12558d3b2edf3e47b4bc0916fa5a2f5330867a98cb7 | | UltraFace-RFB-320 | 1.1
1.2 | PyTorch
ONNX | c722b4427cc71642768baef6e15c659931b56f07425e5d2b0ec033ad41b145b3
8611231d4e6c1f3cda9eb5365518ab1c230df71090c11ed8e701b6ab9b7c58bd | | UltraFace-slim-640 | 1.0
1.4 | PyTorch
ONNX | 02ca778098127c46d2b2680f1c398c7b993c12a424e94c34e6d608beb73481e4
5380b75201e135e4b2c42a5ae4b127e8e474b30f82dfcf8c0cacb254bbf5f243 | | UltraFace-RFB-640 | 1.1
1.5 | PyTorch
ONNX | bf34512b1a93dc234178e8a701ecf25c6afddf335a3226accf62982536e160b5
a3d2fa1ccd444f244716d96fcf0d32d454e422cb8163faa840f80968e25d6f77 |

• Evaluation results on WIDERFace dataset:

| Name | Image Size
(pixels) | Easy | Medium | Hard | |--------------------|------------------------|-------|--------|-------| | UltraFace-slim-320 | 320 x 240 | 0.770 | 0.671 | 0.395 | | UltraFace-RFB-320 | 320 x 240 | 0.787 | 0.698 | 0.438 | | UltraFace-slim-640 | 640 x 480 | 0.853 | 0.819 | 0.539 | | UltraFace-RFB-640 | 640 x 480 | 0.855 | 0.822 | 0.579 |

Inference

shell python detect_img.py --net_type RFB --path imgs/1.jpg

WIDERFace evaluation

• Download WIDERFace validation dataset.
• Move dataset to data/widerface/val directory.

shell cd widerface_evaluate shell python evaluation_on_widerface.py shell python setup.py build_ext --inplace shell python evaluation.py

Export to ONNX format

shell pip install onnx shell python convert_to_onnx.py --model_path models/pretrained/version-slim-320.pth --net_type slim --input_size 320 python convert_to_onnx.py --model_path models/pretrained/version-slim-640.pth --net_type slim --input_size 640 python convert_to_onnx.py --model_path models/pretrained/version-RFB-320.pth --net_type RFB --input_size 320 python convert_to_onnx.py --model_path models/pretrained/version-RFB-640.pth --net_type RFB --input_size 640