boundary-guided-context-aggregation's Introduction

Boundary-Guided-Context-Aggregation

Boundary Guided Context Aggregation for Semantic Segmentation

Haoxiang Ma, Hongyu Yang, Di Huang
In BMVC'2021

Paper

Introduction

This repository is official PyTorch implementation for our BMVC2021 paper. The code is based on semseg

Environments

Anaconda3
Python == 3.7.9
PyTorch == 1.7.1
CUDA ==11.0

Getting Started

Installation

git clone https://github.com/mahaoxiang822/Boundary-Guided-Context-Aggregation.git
cd Boundary-Guided-Context-Aggregation
conda create -n bcanet python=3.7
conda activate bcanet
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0 -c pytorch
pip install -r requirements.txt

Prepare Datasets

For Cityscapes, you can download from Cityscapes

For ADE20K, you can download from ADE20K

You should modify your dataset paths specified in folder config

Train

Download ImageNet pre-trained from GoogleDrive and put them under folder initmodel for weight initialization.
Specify the gpu used in config then do training:

Cityscapes

sh tool/train.sh cityscapes [bcanet50/bcanet101]

ADE20K

To accelerate the training speed on ADE20K, please pre-generate the ground truth of boundary. You can download the pre-generate boundary gt from GoogleDrive

sh tool/trainade.sh ade20k [bcanet50/bcanet101]

Evaluation

Specify the gpu used in config and the checkpoint then do training:
You can download the pre-trained model on cityscapes from GoogleDrive

Validation on Cityscapes

sh tool/test.sh cityscapes [bcanet50/bcanet101]

Test on Cityscapes

sh tool/test.sh cityscapes [bcanet50/bcanet101]

Validation on ADE20K

sh tool/testade.sh ade20k [bcanet50/bcanet101]

Citation

If any part of our paper and repository is helpful to your work, please generously cite with:

@InProceedings{Ma_2021_BMVC,
    author    = {Haoxiang, Ma and Hongyu, Yang and Huang, Di},
    title     = {Boundary Guided Context Aggregation for Semantic Segmentation},
    booktitle = {The British Machine Vision Conference (BMVC)},
    month     = {November},
    year      = {2021}

boundary-guided-context-aggregation's People