Vision DiffMask: Faithful Interpretation of Vision Transformers with Differentiable Patch Masking

Angelos Nalmpantis, Apostolos Panagiotopoulos, John Gkountouras, Konstantinos Papakostas, Wilker Aziz

CVPRW 2023 pp. 3756-3763

doi:10.1109/CVPRW59228.2023.00388 /cvprw/2023/nalmpantis2023cvprw-vision/

Abstract

The lack of interpretability of the Vision Transformer may hinder its use in critical real-world applications despite its effectiveness. To overcome this issue, we propose a post-hoc interpretability method called Vision DiffMask, which uses the activations of the model’s hidden layers to predict the relevant parts of the input that contribute to its final predictions. Our approach uses a gating mechanism to identify the minimal subset of the original input that preserves the predicted distribution over classes. We demonstrate the faithfulness of our method, by introducing a faithfulness task, and comparing it to other state-of-the-art attribution methods on CIFAR-10 and ImageNet-1K, achieving compelling results. To aid reproducibility and further extension of our work, we open source our implementation here.

PDF CVPRW Semantic Scholar

Cite

Text

Nalmpantis et al. "Vision DiffMask: Faithful Interpretation of Vision Transformers with Differentiable Patch Masking." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2023. doi:10.1109/CVPRW59228.2023.00388

Markdown

[Nalmpantis et al. "Vision DiffMask: Faithful Interpretation of Vision Transformers with Differentiable Patch Masking." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2023.](https://mlanthology.org/cvprw/2023/nalmpantis2023cvprw-vision/) doi:10.1109/CVPRW59228.2023.00388

BibTeX

@inproceedings{nalmpantis2023cvprw-vision,
  title     = {{Vision DiffMask: Faithful Interpretation of Vision Transformers with Differentiable Patch Masking}},
  author    = {Nalmpantis, Angelos and Panagiotopoulos, Apostolos and Gkountouras, John and Papakostas, Konstantinos and Aziz, Wilker},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  year      = {2023},
  pages     = {3756-3763},
  doi       = {10.1109/CVPRW59228.2023.00388},
  url       = {https://mlanthology.org/cvprw/2023/nalmpantis2023cvprw-vision/}
}