AMR
AMR copied to clipboard
Activation Modulation and Recalibration Scheme for Weakly Supervised Semantic Segmentation
This repository contains the pytorch codes and trained models described in the AAAI2022 paper "Activation Modulation and Recalibration Scheme for Weakly Supervised Semantic Segmentation".
Authors: Jie Qin, Jie Wu, Xuefeng Xiao, Lujun Li, Xingang Wang.
Overview
Performance
Prerequisites
- Linux
- python 3.8
- pytorch 1.8.1
- torchvision 0.9.1
- NVIDIA GPU
- CUDA 11.1
Preparation
Dependencies
pip install -r requirements.txt
Dataset
Download PASCAL VOC 2012 follwing instructions in http://host.robots.ox.ac.uk/pascal/VOC/voc2012/#devkit.
Pretrained model
The pretrained model can be download in models.
Run
python run_sample.py --train_amr_pass True --make_cam_pass True --eval_cam_pass True
Eval
python run_sample.py --make_cam_pass True --eval_cam_pass True
Acknowledgement
We would like to thank the irn and the AdvCAM for their open-source projects.