saurav-31 / vitol

ViTOL

ViTOL: Vision Transformers for Weakly Supervised Object Localization

Official implementation of the paper ViTOL: Vision Transformer forWeakly Supervised Object Localization which is accepted as CVPRW-2022 paper for L3DIVU-2022.

This repository contains inference code and pre-trained model weights for our model in Pytorch framework. Code is trained and tested in Python 3.6.9 and Pytorch version 1.7.1+cu101

ViTOL-GAR Localization maps:

Model Zoo

We provide pre-trained weights for VITOL with DeiT-S and DeiT-B backbone on ImageNet-1k and CUB datasets below.

ImageNet: ViTOL-base, ViTOL-small

CUB: Updating soon

Results on ImageNet-1k dataset

Method	MaxBoxAccV2	Top1Acc	IOU50	Top1Cls
ViTOL-GAR Small	69.61	54.74	71.86	71.84
ViTOL-LRP Small	68.23	53.62	70.48	71.84
ViTOL-GAR Base	69.17	57.62	71.32	77.08
ViTOL-LRP Base	70.47	58.64	72.51	77.08

Results on CUB dataset

updating soon

Usage

Clone the repository git clone https://github.com/Saurav-31/ViTOL.git

Setup conda environment

conda env create -f environment.yml
conda activate vitol

Dataset preparation

Please refer here for dataset preparation

Inference results on ImageNet

Edit the config files under configs folder

1. Add paths to ImageNet dataset

--data_root=\PATH\TO\DATASET
--metadata_root=\PATH\TO\GROUND_TRUTH 

2. Download ViTOL weights and copy to directory named "pretrained_weights"

--CHECKPOINT_NAME=$VITOL_WEIGHTS_TAR_FILENAME

RUN ViTOL Base with GAR

bash evaluate.sh configs/ilsvrc/ViTOL_GAR_base.yml

RUN ViTOL Small with GAR

bash evaluate.sh configs/ilsvrc/ViTOL_GAR_small.yml

To do

• Setup Training Code for the same
• Train the model with more stronger backbones
• Jupyter notebook for visualization

We borrow code from

_{Evaluating Weakly Supervised Object Localization Methods Right (CVPR 2020) Transformer Interpretability Beyond Attention Visualization (CVPR 2021)}

Contacts

If you have any question about our work or this repository, please don't hesitate to contact us by emails.

• [email protected]
• [email protected]
• [email protected]
• [email protected]

Citation

If you find this work useful, please cite as follows:

@inproceedings{gupta2022vitol,
  title={ViTOL: Vision Transformer for Weakly Supervised Object Localization},
  author={Gupta, Saurav and Lakhotia, Sourav and Rawat, Abhay and Tallamraju, Rahul},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4101--4110},
  year={2022}
}

vitol's Issues

Code for training

Thank you for sharing this awesome work!

Are you planning to publish the code of training soon?

Thanks.!

how to compute the Evaluation Metric "Top1-Loc" and "GT-Known"?