Adversarial Knowledge Transfer from Unlabeled Data

This repository is the official implementation our paper titled "Adversarial Knowledge Transfer from Unlabeled Data" accepted to ACM-MM 2020. Please refer project page for more details.

Implementation Details

Our implementation is in PyTorch [1] with python 3.6.7. We train all our model on GeForce RTX 2080 Ti GPUs. This implementation currently uses one gpu and can be modified to use multiple gpus for larger batch size.

Python Packages

Please refer to the requirements.txt file for all the packages we used to create the environment for training our models. We create an environment in anaconda.

Datasets

This is a working code of our proposed method for PASCAL-VOC/ImageNet experiment. We use PASCAL-VOC[2] dataset as the labeled target dataset and ImageNet[3] as unlabeled source dataset.

Usage

To train a model on PASCAL-VOC and ImageNet experiment WITH GPU

python train.py --pascal_path <path-to-pascal-voc-dataset> \
                --imgnet_path <path-to-imagenet-dataset>   \
                --gpu <gpu-id-to-use>

python train.py --pascal_path /datasets/pascal-voc-2007/   \
                --imgnet_path /datasets/imagenet-dataset/  \
                --gpu 0

To test your trained model WITH GPU

python train.py --pascal_path <path-to-pascal-voc-dataset> \
                --model <path-to-trained-model>            \
                --gpu <gpu-id-to-use>                      \
                --test 1

python train.py --pascal_path /datasets/pascal-voc-2007/   \
                --model ./checkpoints/best-model.pth       \
                --gpu 0                                    \
                --test 1

To train a model on PASCAL-VOC and ImageNet experiment WITHOUT GPU

python train.py --pascal_path <path-to-pascal-voc-dataset> \
                --imgnet_path <path-to-imagenet-dataset>                      

python train.py --pascal_path /datasets/pascal-voc-2007/   \
                --imgnet_path /datasets/imagenet-dataset/   

To test your trained model WITHOUT GPU

python train.py --pascal_path <path-to-pascal-voc-dataset> \
                --model <path-to-trained-model>            \
                --test 1

python train.py --pascal_path /datasets/pascal-voc-2007/   \
                --model ./checkpoints/best-model.pth       \
                --test 1

Citation

@inproceedings{gupta2020adversarial,
  title={Adversarial Knowledge Transfer from Unlabeled Data},
  author={Gupta, Akash and Panda, Rameswar and Paul, Sujoy and Zhang, Jianming and Roy-Chowdhury, Amit K},
  booktitle={Proceedings of the 28th ACM International Conference on Multimedia},
  pages={2175--2183},
  year={2020}
}

Contact

Please contact the first author Akash Gupta ([email protected]) for any questions.

References

1. Paszke, Adam, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. "Automatic differentiation in pytorch." (2017).
   
2. Everingham, Mark, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. "The pascal visual object classes (voc) challenge." International journal of computer vision 88, no. 2 (2010): 303-338.
   
3. Deng, Jia, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. "Imagenet: A large-scale hierarchical image database." In 2009 IEEE conference on computer vision and pattern recognition, pp. 248-255. Ieee, 2009.