https://github.com/hila-chefer/robustvit
[NeurIPS 2022] Official PyTorch implementation of Optimizing Relevance Maps of Vision Transformers Improves Robustness. This code allows to finetune the explainability maps of Vision Transformers to enhance robustness.
https://github.com/hila-chefer/robustvit
explainability neurips neurips-2022 robustness vision-transformer
Last synced: 2 days ago
JSON representation
[NeurIPS 2022] Official PyTorch implementation of Optimizing Relevance Maps of Vision Transformers Improves Robustness. This code allows to finetune the explainability maps of Vision Transformers to enhance robustness.
- Host: GitHub
- URL: https://github.com/hila-chefer/robustvit
- Owner: hila-chefer
- Created: 2022-05-26T11:11:08.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2022-11-22T08:57:48.000Z (almost 3 years ago)
- Last Synced: 2025-07-31T11:36:35.146Z (3 months ago)
- Topics: explainability, neurips, neurips-2022, robustness, vision-transformer
- Language: Jupyter Notebook
- Homepage:
- Size: 16.6 MB
- Stars: 131
- Watchers: 4
- Forks: 13
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Official PyTorch implementation of **[Optimizing Relevance Maps of Vision Transformers Improves Robustness](https://arxiv.org/abs/2206.01161) [NeurIPS 2022]**
This code allows to finetune the explainability maps of Vision Transformers to enhance robustness.
## HuggingFace space + Colab notebook to run examples of the finetuned vs the original models:
[](https://colab.research.google.com/github/hila-chefer/RobustViT/blob/master/RobustViT.ipynb)[](https://huggingface.co/spaces/Hila/RobustViT)
[](https://www.youtube.com/watch?v=i_bY-IDyPD8)
## Updates:
06/05/2022 **Added a [HuggingFace Spaces demo](https://huggingface.co/spaces/Hila/RobustViT)**:
## Method overview:
The method employs loss functions directly to the explainability maps to ensure that the model is focused mostly on the foreground of the image:
Using a short finetuning process with only 3 labeled examples from 500 classes, our method improves robustness of ViT models across different model sizes and training techniques, even when data augmentations/ regularization are applied.
## Model zoo
Below are links to download finetuned models for the base models of [ViT AugReg](https://arxiv.org/abs/2106.10270) (this is also the model that appears on [timm](https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py)), vanilla ViT, and DeiT.
These are also the weights used in our [colab notebook](https://colab.research.google.com/github/hila-chefer/RobustViT/blob/master/RobustViT.ipynb).
| Path | Description
| :--- | :----------
|[AugReg-B](https://drive.google.com/file/d/1jbWiuBrL4sKpAjG3x4oGbs3WOC2UdbIb/view?usp=sharing) | Finetuned ViT Augreg base model.
|[ViT-B](https://drive.google.com/file/d/1vDmuvbdLbYVAqWz6yVM4vT1Wdzt8KV-g/view?usp=sharing) | Finetuned vanilla ViT base model.
|[DeiT-B](https://drive.google.com/file/d/1DHKX_s8rVCDiX4pwnuCCZdGWsOl4SFMn/view?usp=sharing)| Finetuned DeiT base model.
## Requirements
* `pytorch==1.7.1`
* `torchvision==0.8.2`
* `timm==0.4.12`
## Producing Segmentation Data
### Using ImageNet-S
To use the ImageNet-S labeled data, [download the `ImageNetS919` dataset](https://github.com/UnsupervisedSemanticSegmentation/ImageNet-S)
### Using TokenCut for unsupervised segmentation
1. Clone the TokenCut project
```
git clone https://github.com/YangtaoWANG95/TokenCut.git
```
2. Install the dependencies
Python 3.7, PyTorch 1.7.1, and CUDA 11.2. Please refer to the official installation. If CUDA 10.2 has been properly installed:
```
pip install torch==1.7.1 torchvision==0.8.2
```
Followed by:
```
pip install -r TokenCut/requirements.txt
3. Use the following command to extract the segmentation maps:
```
python tokencut_generate_segmentation.py --img_path --out_dir
```
## Finetuning ViT models
To finetune a pretrained ViT model use the `imagenet_finetune.py` script. Notice to uncomment the import line containing the pretrained model you
wish to finetune.
Usage example:
```bash
python imagenet_finetune.py --seg_data --data --gpu 0 --lr --lambda_seg --lambda_acc --lambda_background --lambda_foreground
```
Notes:
* For all models we use :
* `lambda_seg=0.8`
* `lambda_acc=0.2`
* `lambda_background=2`
* `lambda_foreground=0.3`
* For **DeiT** models, a temperature is required as follows:
* `temperature=0.65` for DeiT-B
* `temperature=0.55` for DeiT-S
* The learning rates per model are:
* ViT-B: 3e-6
* ViT-L: 9e-7
* AR-S: 2e-6
* AR-B: 6e-7
* AR-L: 9e-7
* DeiT-S: 1e-6
* DeiT-B: 8e-7
## Baseline methods
Notice to uncomment the import line containing the pretrained model you wish to finetune in the code.
### GradMask
Run the following command:
```bash
python imagenet_finetune_gradmask.py --seg_data --data --gpu 0 --lr --lambda_seg --lambda_acc
```
All hyperparameters for the different models can be found in section D of the supplementary material.
### Right for the Right Reasons
Run the following command:
```bash
python imagenet_finetune_rrr.py --seg_data --data --gpu 0 --lr --lambda_seg --lambda_acc
```
All hyperparameters for the different models can be found in section D of the supplementary material.
## Evaluation
### Robustness Evaluation
1. Download the evaluation datasets:
* [INet-A](https://github.com/hendrycks/natural-adv-examples)
* [INet-R](https://github.com/hendrycks/imagenet-r)
* [INet-v2](https://github.com/modestyachts/ImageNetV2)
* [ObjectNet](https://objectnet.dev/)
* [SI-Score](https://github.com/google-research/si-score)
2. Run the following script to evaluate:
```bash
python imagenet_eval_robustness.py --data --batch-size --evaluate --checkpoint
```
* Notice to uncomment the import line containing the pretrained model you wish to evaluate in the code.
* To evaluate the original model simply omit the `checkpoint` parameter.
* For the INet-v2 dataset add `--isV2`.
* For the ObjectNet dataset add `--isObjectNet`.
* For the SI datasets add `--isSI`.
### Segmentation Evaluation
Our segmentation tests are based on the test in the official implementation of [Transformer Interpretability Beyond Attention Visualization](https://github.com/hila-chefer/Transformer-Explainability).
1. [Download the ImageNet segmentation test set](https://github.com/hila-chefer/Transformer-Explainability#section-a-segmentation-results).
2. Run the following script to evaluate:
```bash
PYTHONPATH=./:$PYTHONPATH python SegmentationTest/imagenet_seg_eval.py --imagenet-seg-path
```
* Notice to uncomment the import line containing the pretrained model you wish to evaluate in the code.
### Credits
* The TokenCut code is built on top of [LOST](https://github.com/valeoai/LOST), [DINO](https://github.com/facebookresearch/dino), [Segswap](https://github.com/XiSHEN0220/SegSwap), and [Bilateral_Sovlver](https://github.com/poolio/bilateral_solver).
* Our ViT code is based on the [pytorch-image-models](https://github.com/rwightman/pytorch-image-models) repository.
* Our ImageNet finetuning code is based on [code from the official PyTorch repo](https://github.com/pytorch/examples/blob/main/imagenet/main.py).
* The code to convert ObjectNet classes to ImageNet classes was taken from [the torchprune repo](https://github.com/lucaslie/torchprune/blob/b753745b773c3ed259bf819d193ce8573d89efbb/src/torchprune/torchprune/util/datasets/objectnet.py).
* The code to convert SI-Score classes to ImageNet classes was taken from [the official implementation](https://github.com/google-research/si-score).
We would like to sincerely thank the authors for their great works.
## Citing our paper
If you make use of our work, please cite our paper:
```
@inproceedings{
chefer2022optimizing,
title={Optimizing Relevance Maps of Vision Transformers Improves Robustness},
author={Hila Chefer and Idan Schwartz and Lior Wolf},
booktitle={Thirty-Sixth Conference on Neural Information Processing Systems},
year={2022},
url={https://openreview.net/forum?id=upuYKQiyxa_}
}
```