Exploring Intrinsic Discrimination and Consistency for Weakly Supervised Object Localization

Official PyTorch implementation of ''Exploring Intrinsic Discrimination and Consistency for Weakly Supervised Object Localization'' (IDC).

To Do

• [x] Evaluation code and pretrained models for IDC.
• [x] Training code.

License

Our code is released under the Creative Commons BY-NC-SA 3.0 (see LICENSE for more details), available only for non-commercial use.

Requirements

• python 3.6
• scipy 1.7.3
• torch 1.11.0
• torchvision 0.12.0
• opencv-python 4.6.0.66
• PyYAML 6.0
• scikit-image 0.19.3
• Pillow 9.2.0

Get started

Start

git clone https://github.com/vignywang/IDC.git
cd IDC-master

Download Datasets

• CUB (http://www.vision.caltech.edu/visipedia/CUB-200-2011.html)
• CUB-200-2011 dataset tree structure after preprocessing.*

  datasets
  └───CUB_200_2011
  └───train
  └───test

  Download Pretrained Models

Training

python train.py  --gpu 0 

Evaluation

python evaluation.py --config ${congfigs/vgg_CUB.yaml} --gpu 0 --epoch 29

The evaluation results will be displayed as:

Cls-Loc acc 0.7236417930523903
Cls-Loc acc Top 5 0.8606499876562411
GT Loc acc 0.9335930243796378
MaxBoxAccV2 acc 0.8715766827518584

Acknowledgements

Part of our evaluation and training code based on PSOL (CVPR2020), BAS (CVPR2022) and CREAM (CVPR2022).

Thanks for their works and sharing.