Skip to content

Identifying salient image regions in Deep Neural Network Classifications.

Notifications You must be signed in to change notification settings

dwil2444/DNN_Attention

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Generate saliency maps for model classifications of fashion-mnist images.

Uses the RISE technique outlined here.

Reading List

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.

About

Identifying salient image regions in Deep Neural Network Classifications.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published