Facial Landmark Detection

1. Quick start

（1）Clone the project

git clone https://github.com/HRNet/HRNet-Facial-Landmark-Detection.git

（2）Install dependencies

pip3 install -r requirements.txt

（3） download pre-trained model and test

python3 camera.py--cfg <CONFIG-FILE> --model-file <MODEL WEIGHT> 
# example:
python3 camera.py  --cfg experiments/face_landmark_detection_wflw_shufflenet_large.yaml  --model-file pretrained/shufflenet_plus.pth

2. Datasets

• Download the annotations files from: Google Drive: https://drive.google.com/file/d/1W8p0MWmUtWxH1B2LsImtg3JVO-o5AU9u/view?usp=sharing

  Baiduyu Link: https://pan.baidu.com/s/10l70jaoWf5ls4t6HMoFk1w Access Code: 6ipb

• Download images (WFLW) from official websites and then put them into images folder for each dataset.

Your data directory should look like this:

.
└──data
    └── wflw
        ├── face_landmarks_wflw_test_blur.csv
        ├── face_landmarks_wflw_test.csv
        ├── face_landmarks_wflw_test_expression.csv
        ├── face_landmarks_wflw_test_illumination.csv
        ├── face_landmarks_wflw_test_largepose.csv
        ├── face_landmarks_wflw_test_makeup.csv
        ├── face_landmarks_wflw_test_occlusion.csv
        ├── face_landmarks_wflw_train.csv
        └── images

2 directories, 8 files

3. Training

python train.py --cfg <CONFIG-FILE>
# example:
python3 train.py --cfg experiments/face_alignment_wflw_hrnet_w18.yaml

4. benchmark

WFLW

NME	model_size	test	pose	illumination	occlution	blur	makeup	expression
shufflenet_plus	13.8M	4.79	8.56	4.73	5.80	5.47	4.77	5.15
HRNet	39.2M	4.60	7.86	4.57	5.42	5.36	4.26	4.78

5. project structure

.
├── data
│   └── wflw
│       ├── face_landmarks_wflw_test_blur.csv
│       ├── ...
│       ├── face_landmarks_wflw_train.csv
│       └── images
├── experiments
│   └── face_landmark_detection_wflw_shufflenet_large.yaml
├── output
│   ├── log
│   │   └── WFLW
│   └── WFLW
│       └── face_landmark_detection_wflw_shufflenet_large
├── README.md
├── requirements.txt
├── src
│   ├── datasets.py
│   ├── __init__.py
│   ├── loss.py
│   ├── models
│   │   ├── hrnet.py
│   │   ├── __init__.py
│   │   ├── shufflenet_bak.py
│   │   ├── shufflenet.py
│   │   └── utils.py
│   ├── transforms.py
│   └── utils.py
├── test.py
└── train.py  

13 directories, 28 files

6. TBD

• face pose weighted
• heatmap
• graph network
• model、dataset、loss
• deployment
• video stable

7. Reference

https://github.com/HRNet/HRNet-Facial-Landmark-Detection