Generate saliency maps for model classifications of fashion-mnist images.
Uses the RISE technique outlined here.
| Title | Author | Conf | Notes | Link |
|---|---|---|---|---|
| RISE: Randomized Input Sampling for Explanation of Black-box Models. | V Petsiuk, A Das, K Saenko | BMVC 2018 | RISE: Saliency Technique for Blackbox models | http://arxiv.org/abs/1806.07421 |
| Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization. | RR Selvaraju, M Cogswell, A Das, R Vedantam, D Parikh, D Batr | ICCV 2017 | GRAD-CAM technique for saliency which tracks gradient changes by sampling feature maps. | https://arxiv.org/pdf/1610.02391.pdf |
| On Guiding Visual Attention with Language Specification | S Petryk, L Dunlap, K Nasseri, J Gonzalez, T Darrell, A Rohrbach | arXiv preprint arXiv:2202.08926 | Training for Saliency with augmented loss functions. | https://arxiv.org/pdf/2202.08926.pdf |
| "Why Should I Trust You?": Explaining the Predictions of Any Classifier | Marco Tulio Ribeiro, Sameer Singh, Carlos Guestrin | KDD2016 | Learns an interpretable model locally around the prediction | https://arxiv.org/pdf/1602.04938.pdf |
Run setup.sh to install the dependencies required.
The train.py script will train a CNN for classification on Fashion-MNIST.
The rise_mnist.py script will produce a sample saliency map given a single example from the Fashion-MNIST validation set. (Everything is included in the rise_mnist.ipynb notebook.