Reimplement RetinaFace with Pytorch
$ git clone https://github.com/supernotman/RetinaFace_Pytorch.git
$ cd RetinaFace_Pytorch/
$ sudo pip install -r requirements.txt
Pytorch version 1.1.0+ and torchvision 0.3.0+ are needed.
-
Download widerface dataset
-
Download annotations (face bounding boxes & five facial landmarks) from baidu cloud or dropbox
-
Organise the dataset directory as follows:
widerface/
train/
images/
label.txt
val/
images/
label.txt
test/
images/
label.txt
$ train.py [-h] [data_path DATA_PATH] [--batch BATCH]
[--epochs EPOCHS]
[--shuffle SHUFFLE] [img_size IMG_SIZE]
[--verbose VERBOSE] [--save_step SAVE_STEP]
[--eval_step EVAL_STEP]
[--save_path SAVE_PATH]
[--depth DEPTH]
For multi-gpus training, run:
$ CUDA_VISIBLE_DEVICES=0,1,2,3,4,5 python train.py --data_path /widerface --batch 32 --save_path ./out
---- [Epoch 39/200, Batch 400/403] ----
+----------------+-----------------------+
| loss name | value |
+----------------+-----------------------+
| total_loss | 0.09969855844974518 |
| classification | 0.09288528561592102 |
| bbox | 0.0034053439740091562 |
| landmarks | 0.003407923271879554 |
+----------------+-----------------------+
-------- RetinaFace Pytorch --------
Evaluating epoch 39
Recall: 0.7432201780921814
Precision: 0.906913273261629
You can download the model from baidu cloud or dropbox
$ python detect.py --model_path model.pt --image_path 4.jpg
$ python video_detect.py --model_path model.pt
Hey, I find something interesting and add it in the code. Pose detection Hopenet: https://github.com/natanielruiz/deep-head-pose Now you can estimate pose with RetinaFace and Hopenet. Download pose model
$ python pose_detect.py --f_model model.pt --p_model hopenet.pkl --image_path test.jpg
also you can detect in video
$ python pose_detect.py --f_model model.pt --p_model hopenet.pkl --type video --video_path test.avi
- Wider Face mAP calculation
- Deformable Convolution
- More models support
- Random crop and color distortion
- Graph Convolution
- Bug fix