soka

Collaboration by competition: Self-coordinated knowledge amalgamation for multi-talent student learning

Note: This part of the code is a reproduction of the paper by a junior student in our lab, and it may differ from the original paper in some details. Therefore, please refer to the original paper for the specific details.

0. Preparation

Taskonomy Dataset should be downloaded before training

Then, you can use the script */examples/knowledge_amalgamation/soka/soka_utils/get_features_te.py to obtain the soft_target and encoder_features of teacher models.

python get_features_te.py --tasks normal/depth_euclidean/edge_occlusion/edge_texture

1. train

You can get the student model by the following step:

python soka_tasks.py 

The trained checkpoints will be saved in the ./checkpoints/xx.pth

2. evaluation

python soka_tasks.py  --test_only --ckpt ./checkpoints/xx.pth

Citation

If you found this work useful for your research, please cite our paper:

@inproceedings{luo2020collaboration,
  title={Collaboration by competition: Self-coordinated knowledge amalgamation for multi-talent student learning},
  author={Luo, Sihui and Pan, Wenwen and Wang, Xinchao and Wang, Dazhou and Tang, Haihong and Song, Mingli},
  booktitle={Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part VI 16},
  pages={631--646},
  year={2020},
  organization={Springer}
}