Implementation of the paper : Real Time Image Saliency for Black Box Classifiers https://arxiv.org/abs/1705.07857
| Example 1 | Example 2 |
|---|---|
 |
 |
 |
 |
 |
 |
NIPS 2017 Paper Implementation Challenge . Even though the author's implementation exists, this is an attempt to make a more user friendly version of the code and especially purposed as a learning tool for the models in the paper.
Differences from the Official Implementation:
- Made for training on the Cifar10 dataset, official repo is meant to train the ImageNet Dataset
- Less Verbose, only the core details present. Loss, Model and Trainer. Readable Code
- model.py -> Main model hosted
- resnet.py -> Black Box Classifier
- train_classifier -> Trains classifier Model
- train_saliency -> Trains Saliency Model
- Logging
- Responsive Training bar ala Keras
- Validation Training
- Make code to work for any dataset
- Add circle ci and Docker support
- Command Line Interface
- ImageNet training
Referred to the author's repo ,this project was made on the shoulders of giants. Official Repo: https://github.com/PiotrDabkowski/pytorch-saliency