CEASC: Adaptive Sparse Convolutional Networks with Global Context Enhancement for Faster Object Detection on Drone Images
The repo is the official implementation of CEASC.
Our CEASC module is at mmdet/models/dense_heads
Our Sparse Convolution Implementation is at Sparse_conv
Our config file is at configs/UAV
Please follow docs/en/get_started.md and install the mmdetection toolbox.
a. Install Pytorch 1.10.1
b. Install MMDetection toolbox, required mmdet >= 2.7.0, mmcv-full >= 1.4.2.
- Our project utilizes mmdet == 2.24.1, mmcv-full == 1.5.1
c. Install albumentations and other packages.
pip install nltk
pip install -r requirements/albu.txtd. Install our Sparse Convolution Implementation
cd ./Sparse_conv
python setup.py install
cd ..You could download VisDrone and UAVDT dataset (COCO Format) from official links or from other repositories like UFPMP-Det.
% training on a single GPU
python tools/train.py /path/to/config-file --work-dir /path/to/work-dir
% training on multi GPUs
bash tools/dist_train.sh /path/to/config-file num-gpus --work-dir /path/to/work-dirWe provide the following checkpoints:
- GFL v1 baseline, corresponding to baseline_gfl_res18_visdrone: Google Drive
- GFL v1 CEASC, corresponding to dynamic_gfl_res18_visdrone: Google Drive
- RetinaNet baseline, corresponding to baseline_retinanet_res18_visdrone: Google Drive
- RetinaNet CEASC, corresponding to dynamic_retinanet_res18_visdrone: Google Drive
python tools/test.py /path/to/config-file /path/to/work-dir/latest.pth --eval bboxIf you find our paper or this project helps your research, please kindly consider citing our paper in your publication.
@misc{ceasc,
title={Adaptive Sparse Convolutional Networks with Global Context Enhancement for Faster Object Detection on Drone Images},
author={Bowei Du and Yecheng Huang and Jiaxin Chen and Di Huang},
year={2023},
eprint={2303.14488},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent