NUST-Machine-Intelligence-Laboratory/MDBA

Python

Multi-Granularity Denoising and Bidirectional Alignment for Weakly Supervised Semantic Segmentation

Introduction

This is the source code for our paper Multi-Granularity Denoising and Bidirectional Alignment for Weakly Supervised Semantic Segmentation

Network Architecture

The architecture of our proposed approach is as follows

Installation

Install PyTorch 1.6 with Python 3 and CUDA 10.1
Clone this repo

git clone https://github.com/NUST-Machine-Intelligence-Laboratory/MDBA.git

Download PASCAL VOC 2012

Download PASCAL VOC 2012 and put it into ./dataset to obtain ./dataset/VOC2012

Testing

Download our trained model VOC_69.5.pth and put it in the ./model folder

python evaluate.py --restore-from ./model/VOC_69.5.pth

python evaluate_crf.py

Training

Download the COCO pre-trained parameters and put it in the ./model folder
Download saliency-generated pseudo labels sal2gt and put it into ./dataset to obtain ./dataset/sal2gt.
Train the segmentation model

python train.py

Testing two-step result

Download our trained model checkpoint_72.0.pth and put it in the ./step2/data/models folder

cd step2

python main.py test --config-path configs/voc12.yaml --model-path data/models/checkpoint_72.0.pth

python main.py crf --config-path configs/voc12.yaml

Two-step training

Generate pseudo labels from the segmentation network trained by our proposed approach.

python gen_label.py --restore-from ./model/VOC_69.5.pth

python gen_label_crf.py 

python gen_label_mining.py

Download the COCO pre-trained parameters for two-step training and put it in the step2/data/models folder. Then train the segmentation model with pseudo labels . (You need to step2/libs/datasets/voc.py set the path for pseudo labels.)

python main.py train --config-path configs/voc12.yaml