Pytorch Face Landmark Detection

Implementation of face landmark detection with PyTorch. The model was trained using coordinate-based regression methods.

  • Support 68-point and 39-point landmark inference.
  • Support different backbone networks.
  • Support ONNX inference.

Inference

Test on a sample folder and save the landmark detection results.

python3 -W ignore test_batch_mtcnn.py

Optimize with ONNX and test on a camera. The pytorch model has been converted to ONNX for fast inference.

python3 -W ignore test_camera_mtcnn_onnx.py

Benchmark Results on 300W

  • Inter-ocular Normalization (ION)
Algorithms Common Challenge Full Set Param # (M) CPU Inference (s)
ResNet18 (224×224) 3.73 7.14 4.39 11.76 /
Res2Net50_v1b (224×224) 3.43 6.77 4.07 26.00 /
Res2Net50_v1b_SE (224×224) 3.37 6.67 4.01 27.05 /
Res2Net50_v1b_ExternalData (224×224) 3.30 5.92 3.81 26.00 /
HRNet_w18_small_v2 (224×224) 3.57 6.85 4.20 13.83 /
MobileNetV2 (224×224) 3.70 7.27 4.39 3.74 1.2
MobileNetV2_SE (224×224) 3.63 7.01 4.28 4.15 /
MobileNetV2 (56×56) 4.50 8.50 5.27 3.74 0.01 (onnx)
MobileNetV2_ExternalData (224×224) 3.48 6.0 3.96 3.74 1.2

Visualization Results

  • Face alignment on 300W dataset
    300w

  • Semi-frontal face alignment on Menpo dataset
    menpo_semi_frontal

  • Profile face alignment on Menpo dataset
    menpo_profile

TODO

The following features will be added soon.

  • Still to come:
    • [x] Support for the 39-point detection
    • [ ] Support for the 106 point detection
    • [ ] Support for heatmap-based inferences

Datasets:

References:

GitHub