Crossmodal knowledge distillation hashing（CKDH）

Install

Our work follows the relevant configuration requirements of ViLBERT, please read the README.md of ViLBERT for more information.

DataSets

• Download the datasets. for example:MIRFLICKR and COCO
• For the method of extracting picture features, please refer to ViLBERT's Data Setup.

Train

• Finetune the network with the classification task to accelerate training convergence. Such as:
  

python pre_teacher_main.py --from_pretrained <pretrained_model_path>

• Teacher NetWork: set the "from_pretrained" and start training.

python train_teacher.py --from_pretrained <pretrained_model_path>

• Student NetWork: set the teacher network model path and start training.

python train_student.py --from_pretrained <teacher_network_model_path>

Citation

If you find this work is helpful in your research, please cite:

@article{wang2022ckdh,
  title={Crossmodal knowledge distillation hashing},
  author={Wang, Jinhui and Jin, Lu and Li, Zechao and Tang, Jinhui},
  journal={SCIENTIA SINICA Technologica},
  year={2022},
  doi={10.1360/SST-2021-0214}
}

Acknowledgement

Thanks to ViLBERT for their help in this work.