Home

Awesome

Convolutional Dynamic Alignment Networks for Interpretable Classifications

Official implementation of the CVPR 2021 paper (oral): Arxiv Paper | GitHub Pages

M. Böhle, M. Fritz, B. Schiele. Convolutional Dynamic Alignment Networks for Interpretable Classifications. CVPR, 2021.

<div style="displaystyle=block;align=center;"><p align="center" > <img src="docs/media/example_figure.png"/> </p> </div>

Overview

Comparison to post-hoc explanation methods evaluated on the CoDA-Nets

<div style="displaystyle=block;align=center;"><p align="center" > <img src="docs/media/comparisons.png"/> </p> </div>

Evaluated on videos

In order to highlight the stability of the contribution-based explanations of the CoDA-Nets, we present some examples for which the output for the respective class of the CoDA-Net was linearly decomposed frame by frame; for more information, see interpretability/eval_on_videos.py.

<div style="displaystyle=block;align=center"><p align="center"> <img width="240px" height="auto" src="docs/media/lorikeet.gif?raw=true"/> <img width="240px" height="auto" src="docs/media/drake2_atts.gif?raw=true"/> <img width="240px" height="auto" src="docs/media/birds_atts.gif?raw=true"/> <img width="240px" height="auto" src="docs/media/goldfinch.gif?raw=true"/> </p></div>

Quantitative Interpretability results

In order to reproduce these plots, check out the jupyter notebook CoDA-Networks Examples. For more information, see the paper and check out the code at interpretability/

<p align="center">Localisation metric</p><p align="center">Pixel removal metric</p>
<p align="center">Compared to others</p>Contributions per LayerContributions per Layer
<p align="center">Trained w/ different temperatures</p>Contributions per LayerContributions per Layer

Copyright and license

Copyright (c) 2021 Moritz Böhle, Max-Planck-Gesellschaft

This code is licensed under the BSD License 2.0, see license.

Further, you use any of the code in this repository for your research, please cite as:

  @inproceedings{Boehle2021CVPR,
          author    = {Moritz Böhle and Mario Fritz and Bernt Schiele},
          title     = {Convolutional Dynamic Alignment Networks for Interpretable Classifications},
          journal   = {IEEE/CVF Conference on Computer Vision and Pattern Recognition ({CVPR})},
          year      = {2021}
      }