Skip to content

Linear Gaussian Bounding Box Representation and Ring-Shaped Rotated Convolution for Oriented Object Detection

Notifications You must be signed in to change notification settings

zhen6618/RotaYolo

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

93 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Linear Gaussian Bounding Box Representation and Ring-Shaped Rotated Convolution for Oriented Object Detection (accepted by Pattern Recognition 2024)

In oriented object detection, current representations of oriented bounding boxes (OBBs) often suffer from the boundary discontinuity problem. Methods of designing continuous regression losses do not essentially solve this problem. Although Gaussian bounding box (GBB) representation avoids this problem, directly regressing GBB is susceptible to numerical instability. We propose linear GBB (LGBB), a novel OBB representation. By linearly transforming the elements of GBB, LGBB avoids the boundary discontinuity problem and has high numerical stability. In addition, existing convolution-based rotation-sensitive feature extraction methods only have local receptive fields, resulting in slow feature aggregation. We propose ring-shaped rotated convolution (RRC), which adaptively rotates feature maps to arbitrary orientations to extract rotation-sensitive features under a ring-shaped receptive field, rapidly aggregating features and contextual information. Experimental results demonstrate that LGBB and RRC achieve state-of-the-art performance. Furthermore, integrating LGBB and RRC into various models effectively improves detection accuracy.

Realted Work

  1. Comparison with Current OBB Representations
  1. Comparison with Current Convolution-Based Rotation-Sensitive Feature Extraction Methods

Methods

  1. Overview
  1. LGBB
  1. RRC

Experiments

  1. Comparison with Current OBB Representations
  1. Comparison with Current Convolution-Based Rotation-Sensitive Feature Extraction Methods
  1. Comparison with Current Oriented Object Detectors

Installation

Refer to both yolov7 and mmrotate

Prepare Your Dataset

DOTA
HRSC2016

Training

# Single GPU training
python train.py --workers 8 --device 0 --batch-size 2 --data data/dota.yaml --img 1024 1024 --cfg cfg/training/RotaYolo_RotaConv.yaml --weights '' --hyp data/hyp.scratch.dota.yaml

# Multiple GPU training
python -m torch.distributed.launch --nproc_per_node 4 --master_port 9527 train.py --workers 8 --device 0,1,2,3 --sync-bn --batch-size 8 --data data/dota.yaml --img 1024 1024 --cfg cfg/training/RotaYolo_RotaConv.yaml --weights '' --hyp data/hyp.scratch.dota.yaml

Detecting

python detect.py --weights 'weights/best.pt' --source 'datasets/DOTA/demo.png' --img-size 1024 --conf-thres 0.5 --iou-thres 0.2 --device 0

Citation

@article{ZHOU2024110677,
      title = {Linear Gaussian bounding box representation and ring-shaped rotated convolution for oriented object detection},
      journal = {Pattern Recognition},
      volume = {155},
      pages = {110677},
      year = {2024},
      author = {Zhen Zhou and Yunkai Ma and Junfeng Fan and Zhaoyang Liu and Fengshui Jing and Min Tan},
}

Acknowledgement

mmrotate

yolov7

About

Linear Gaussian Bounding Box Representation and Ring-Shaped Rotated Convolution for Oriented Object Detection

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published