mmrotate-dcfl

Official implementation for the CVPR23 paper: Dynamic Coarse-to-Fine Learning for Oriented Tiny Object Detection. arxiv

Introduction

DCFL is a learning framework for detecting oriented tiny objects.

Installation and Get Started

Required environments:

• Linux
• Python 3.7+
• PyTorch 1.10.0+
• CUDA 9.2+
• GCC 5+
• MMdet 2.23.0+
• MMCV-DCFL

Install: Note that this repository is based on the MMRotate. Assume that your environment has satisfied the above requirements, please follow the following steps for installation.

git clone https://github.com/Chasel-Tsui/mmrotate-dcfl.git
cd mmrotate-dcfl
pip install -r requirements/build.txt
python setup.py develop

Main Results

DOTA-v2.0

Method	Backbone	AP50	Angle	lr schd	Aug	Batch Size	Configs	Speed
RetinaNet-O	ResNet50 (1024,1024,200)	46.68	le135	1x	Flipping	2	retinanet_obb_r50_dota2	20.8 FPS
R3Det w/ KLD	ResNet50 (1024,1024,200)	47.26	le135	1x	Flipping	2	r3det_le135_r50_dota2	16.2 FPS
ATSS-O	ResNet50 (1024,1024,200)	49.57	le135	1x	Flipping	2	atss_le135_r50_dota2	-
S2A-Net	ResNet50 (1024,1024,200)	49.86	le135	1x	Flipping	2	s2a_le135_r50_dota2	18.9 FPS
DCFL	ResNet50 (1024,1024,200)	51.57	le135	1x	Flipping	2	dcfl_r50_dota2	20.9 FPS
DCFL	ResNet101 (1024,1024,200)	52.54	le135	1x	Flipping	2	dcfl_r101_dota2	-

Visualization

Predictions of the RetinaNet-O are shown in the first row, predictions of the DCFL are shown in the second row. Note that the green box denotes the True Positive, the red box denotes the False Negative and the blue box denotes the False Positive predictions.

Citation

If you find this work helpful, please consider citing:

@InProceedings{Xu_2023_CVPR,
    author    = {Xu, Chang and Ding, Jian and Wang, Jinwang and Yang, Wen and Yu, Huai and Yu, Lei and Xia, Gui-Song},
    title     = {Dynamic Coarse-To-Fine Learning for Oriented Tiny Object Detection},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {7318-7328}
}