Revisiting CAM
We prove the CAM method can be simplified to enable end-to-end training. The proof refers to Section 3.1.
We apply two classifiers to discover complementary regions of target objects.
- Python2.7
- PyTorch
- tqdm
- Download the ILSVRC dataset and save them to
$data$
git clone https://github.com/xiaomengyc/ACoL.git
cd ACoL
mkdir snapshots
cd scripts
bash train_vgg_imagenet.sh
If you find this code helpful, please consider to cite this paper:
@inproceedings{zhang2018adversarial,
title={Adversarial complementary learning for weakly supervised object localization},
author={Zhang, Xiaolin and Wei, Yunchao and Feng, Jiashi and Yang, Yi and Huang, Thomas},
booktitle={IEEE CVPR},
year={2018}
}