ROI-Nets pytorch version

Pytorch implementation of Action Unit Detection with Region Adaptation, Multi-labeling Learning and Optimal Temporal Fusing

How to use:

• clone codes
• download vgg pretrained parameters vgg16 in data directory
• prepare labels for every image, the format is shown in data/labels_224_landmarks_sample.txt
• modify parameters in lib/train_test.py, like image_dir,save_dir,person_batch
• run code: cd lib; CUDA_VISIBLE_DEVICES=0 python train_test.py

Result:

1. train loss
   
2. train f1
   
3. result
   

Reference: https://github.com/AlexHex7/DRML_pytorch and https://github.com/wiibrew/EAC-Net