Interpretable Visual Reasoning via Probabilistic Formulation under Natural Supervision

Code release for "Interpretable Visual Reasoning via Probabilistic Formulation under Natural Supervision" (ECCV 2020) (PDF)

More details can reffer to the Supplementary Material.

@article{han2020interpretable,
  title={Interpretable Visual Reasoning via Probabilistic Formulation under Natural Supervision},
  author={Han, Xinzhe and Wang, Shuhui and Su, Chi and Zhang, Weigang and Huang, Qingming and Tian, Qi},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  year={2020}  
}

This repo contains the experiments on both VQA v2 and CLEVR dataset. The implementation details are listed in ./exp_vqa and ./exp_clevr.

Dependencies

• python 3.6
• pytorch >= 1.1
• spaCy 2.2
• tqdm 4.42
• h5py 2.10
• numpy 1.18

Acknowledgements

Baseline models are referred to https://github.com/KaihuaTang/VQA2.0-Recent-Approachs-2018.pytorch

Data process for CLEVR are referred to NS-VQA