VISION DIFFMASK: Interpretability of Computer Vision models with Differentiable Patch Masking
Jun 1, 2023ยท,,,,ยท
1 min read
Angelos Nalmpantis*
Apostolos Panagiotopoulos*
Admin"*"
Konstantinos Papakostas*
Wilker Aziz
Abstract
We developed a competitive post-hoc interpretation method for vision tasks, focusing on both faithfulness and plausibility. The method implements a minimal input subset selection latent variable model for Vision Transformer, preserving the predicted distribution over classes. We trained the explanation model to produce attributions that are interpretable in a formal mathematical sense.
Type
Publication
Conference on Computer Vision and Pattern Recognition XAI4CV Workshop
This work was presented at CVPR 2023 in Vancouver, Canada.
*Equal contribution