DCAM

The official code of JVCIR 2023 paper (DCAM: Disturbed Class Activation Maps for Weakly Supervised Semantic Segmentation).

Citation

inproceedings{2023dcam,
  title={DCAM: Disturbed class activation maps for weakly supervised semantic segmentation},
  author={Lei, Jie and Yang, Guoyu and Wang, Shuaiwei and Feng, Zunlei and Liang, Ronghua},
  journal={Journal of Visual Communication and Image Representation},
  pages={103852},
  year={2023},
  publisher={Elsevier}
}

Prerequisite

• Python 3.6, PyTorch 1.9.0, and others in environment.yml
• You can create the environment from environment.yml file

conda env create -f environment.yml

Usage (PASCAL VOC)

Step 1. Prepare dataset.

• Download PASCAL VOC 2012 devkit from official website. Download.
• You need to specify the path ('voc12_root') of your downloaded devkit in the following steps.

Step 2. Train DCAM and generate seeds.

• Please specify a workspace to save the model and logs.

CUDA_VISIBLE_DEVICES=0 python run_sample.py --voc12_root ./VOCdevkit/VOC2012/ --work_space YOUR_WORK_SPACE --train_cam_pass True --train_dcam_pass True --train_dcam_sce_pass True --make_dcam_sce_pass True --eval_cam_pass True 

Step 3. Train IRN and generate pseudo masks.

CUDA_VISIBLE_DEVICES=0 python run_sample.py --voc12_root ./VOCdevkit/VOC2012/ --work_space YOUR_WORK_SPACE --cam_to_ir_label_pass True --train_irn_pass True --make_sem_seg_pass True --eval_sem_seg_pass True 

Step 4. Train semantic segmentation network.

To train DeepLab-v2, we refer to deeplab-pytorch. To train DeepLab-v3+, we refer to deeplabv3-plus-pytorch. Please replace the groundtruth masks with generated pseudo masks.

Acknowledgment

This code is borrowed from ReCAM, thanks Zhaozheng Chen.