• Stars
    star
    259
  • Rank 157,669 (Top 4 %)
  • Language
    Python
  • License
    Apache License 2.0
  • Created over 5 years ago
  • Updated over 5 years ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Grid R-CNN

An mmdetection based implementation of updated Grid R-CNN published on CVPR 2019. Original paper is here, details of updates can be seen in this technical report.

Installation

This project is based on mmdetection object detection framework.

Requirements

Install Grid R-CNN with mmdetection

a. Create a conda virtual environment and activate it. Then install Cython.

conda create -n open-mmlab python=3.7 -y
source activate open-mmlab

conda install cython

b. Install PyTorch stable or nightly and torchvision following the official instructions.

c. Clone the this repository.

d. Compile cuda extensions.

./compile.sh

e. Setup mmdetection (other dependencies will be installed automatically).

python setup.py develop
# or "pip install -e ."

Dataset Preparation

It is recommended to symlink the dataset root to your project.

Grid_RCNN
β”œβ”€β”€ mmdet
β”œβ”€β”€ tools
β”œβ”€β”€ configs
β”œβ”€β”€ data
β”‚   β”œβ”€β”€ coco
β”‚   β”‚   β”œβ”€β”€ annotations
β”‚   β”‚   β”œβ”€β”€ train2017
β”‚   β”‚   β”œβ”€β”€ val2017
β”‚   β”‚   β”œβ”€β”€ test2017
β”‚   β”œβ”€β”€ VOCdevkit
β”‚   β”‚   β”œβ”€β”€ VOC2007
β”‚   β”‚   β”œβ”€β”€ VOC2012

Training and Testing

We provide training and testing scripts and config files for Grid R-CNN. An example configuration file is here, corresponding checkpoint is here. Note that the batchsize we used is 64(32 GPUs/2 images per GPU), you could adjust the learning rate and warmup schedule according to your batchsize.

Please see GETTING_STARTED.md for more basic usage of mmdetection.

Results

Training on Res50/101 FPN backbone and Testing on COCO minival. Inference speed is tested on our TITANXp GPU cluster.

Method lr sched AP [email protected] [email protected] AP@S AP@M AP@L Inference speed
Res50-FPN 2x 37.4% 59.3% 40.3% 21.8% 40.9% 47.9% 0.09s
Res50-FPN-Grid R-CNN 2x 40.4% 58.6% 43.7% 23.2% 44.2% 52.4% 0.11s
Res101-FPN 2x 39.5% 61.2% 43.1% 22.7% 43.7% 50.8% 0.12s
Res101-FPN-Grid R-CNN 2x 42.0% 60.6% 45.4% 24.1% 46.2% 55.2% 0.13s

Citation

If you use our codebase or models in your research, please cite this project.

@InProceedings{Lu_2019_CVPR,
author = {Lu, Xin and Li, Buyu and Yue, Yuxin and Li, Quanquan and Yan, Junjie},
title = {Grid R-CNN},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}

@article{mmdetection,
  title   = {{MMDetection}: Open MMLab Detection Toolbox and Benchmark},
  author  = {Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li,
             Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang, Dazhi Cheng,
             Chenchen Zhu, Tianheng Cheng, Qijie Zhao, Buyu Li, Xin Lu, Rui Zhu, Yue Wu,
             Jifeng Dai, Jingdong Wang, Jianping Shi, Wanli Ouyang, Chen Change Loy, Dahua Lin},
  journal = {arXiv preprint arXiv:1906.07155},
  year    = {2019}
}

License

Grid R-CNN is released under the Apache 2.0 license.