Skip to content

Latest commit

 

History

History
115 lines (87 loc) · 4.15 KB

README.md

File metadata and controls

115 lines (87 loc) · 4.15 KB
Raw

ContextLab

ContextLab: A Toolbox for Context Feature Augmentation developed with PyTorch

Introduction

The master branch works with PyTorch 1.1 or higher

ContextLab is an open source context feature augmentation toolbox based on PyTorch. It is a part of the Open-PLUS project developed by ShanghaiTech PLUS Lab

Major Features

  • Modular Design

  • High Efficiency

  • State-of-the-art Performance

We have implemented several context augmentation algorithms in PyTorch with comparable performance.

License

This project is released under the MIT License

Updates

V0.2.0 (27/09/2019)

  • Support for CCNet, TreeFilter and EMANet

v0.1.0 (26/07/2019)

  • Start the project

Benchmark and Model Zoo

Method Block-wise Stage-wise Paper
Non-local Network ✗ ✓ CVPR 18
Dual-attention ✗ ✓ CVPR 19
GCNet ✗ ✓ Arxiv
CCNet ✓ ✓ ICCV 19
LatentGNN ✗ ✓ ICML 19
TreeFilter ✗ ✓ NIPS 19
EMANet ✗ ✓ ICCV 19

Installation

git clone https://github.com/SHTUPLUS/contextlab.git
cd contextlab/
python setup.py build develop

Exapmles

# GCNet
from contextlab.layers import GlobalContextBlock2d
# Dual-Attention
from contextlab.layers import SelfAttention
# LatentGNN
from contextlab.layers import LatentGNN
# TreeFilter
from contextlab.layers import MinimumSpanningTree, TreeFilter2D
# CCNet
from contextlab.layers import CrissCrossAttention
# EMAttetnion
from contextlab.layers import EMAttentionUnit

To do

  • Experiments on Segmentation and Detection
  • Performance Comparison

Contributing

We appreciate all contributions to improve ContextLab. Please refer to CONTRIBUTING.md for the contributing guideline.

Acknowledgement

ContextLab is an open source project that is contributed by researchers and engineers from various colledges and companies. We appreciate all the contributors who implement their methods or add new features.

We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to reimplement existing methods and develop their own new segmentation methods.

Citation

@misc{contextlab,
  title   = {{ContextLab}: A Toolbox for Context Feature Augmentation},
  author  = {Songyang Zhang},
  year={2019}
}

Contact

email: sy.zhangbuaa@gmail.com

Related Projects