This repository contains the pytorch codes and trained models described in the AAAI2022 paper "Activation Modulation and Recalibration Scheme for Weakly Supervised Semantic Segmentation". This algorithm is proposed by ByteDance, Intelligent Creation, CV AutoML Team (字节跳动-智能创作-CV AutoML团队).

Authors: Jie Qin, Jie Wu, Xuefeng Xiao, Lujun Li, Xingang Wang.

• Linux
• python 3.8
• pytorch 1.8.1
• torchvision 0.9.1
• NVIDIA GPU
• CUDA 11.1

pip install -r requirements.txt

Download PASCAL VOC 2012 follwing instructions in http://host.robots.ox.ac.uk/pascal/VOC/voc2012/#devkit.

The pretrained model can be download in link. Extraction code: mepu.

python run_sample.py --train_amr_pass True --make_cam_pass True --eval_cam_pass True

python run_sample.py --make_cam_pass True --eval_cam_pass True

We would like to thank the irn and the AdvCAM for their open-source projects.