/Unc-SOD

Primary LanguagePythonApache License 2.0Apache-2.0

Unc-SOD: An uncertainty Learning Framework for Small Object Detection

📢 Introduction

This is the official repository of our paper titled Unc-SOD: An uncertainty Learning Framework for Small Object Detection, and the full implementation will be available after the peer-review.

🎡 Dependencies

  • CUDA 11.3
  • Python 3.8
  • PyTorch 1.10.0
  • TorchVision 0.11.0
  • mmcv-full 1.5.0
  • numpy 1.22.4

📂 Datasets

Our work is based on the large-scale small object detection benchmark SODA, which comprises two sub datasets SODA-D and SODA-A. Download the dataset(s) from corresponding links below.

The data preparation for SODA differs slightly from that of conventional object detection datasets, as it requires the initial step of splitting the original images. Srcipts to obtain sub-images of SODA-D can be found at tools/img_split. For SODA-A, please refer to SODA-mmrotate. More details about SODA please refer to the Dataset Homepage.

🛠️ Install

This repository is build on MMDetection 2.26.0 which can be installed by running the following scripts. Please ensure that all dependencies have been satisfied before setting up the environment.

git clone https://github.com/shaunyuan22/Unc-SOD
cd Unc-SOD
pip install -v -e .

Moreover, please refer to SODA-mmrotate for MMRotate installation if you want to perform evaluation on the SODA-A dataset.

🚀 Training

  • Single GPU:
python ./tools/train.py ${CONFIG_FILE}
  • Multiple GPUs:
bash ./tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM}

📈 Evaluation

  • Single GPU:
python ./tools/test.py ${CONFIG_FILE} ${WORK_DIR} --eval bbox
  • Multiple GPUs:
bash ./tools/dist_test.sh ${CONFIG_FILE} ${WORK_DIR} ${GPU_NUM} --eval bbox

🏆 Result

Result on SODA-D

Method Schedule $AP$ $AP_{50}$ $AP_{75}$ $AP_{eS}$ $AP_{rS}$ $AP_{gS}$ $AP_N$
RetinaNet $1 \times$ 28.2 57.6 23.7 11.9 25.2 34.1 44.2
FCOS $1 \times$ 23.9 49.5 19.9 6.9 19.4 30.9 40.9
RepPoints $1 \times$ 28.0 55.6 24.7 10.1 23.8 35.1 45.3
ATSS $1 \times$ 26.8 55.6 22.1 11.7 23.9 32.2 41.3
YOLOX $70e$ 26.7 53.4 23.0 13.6 25.1 30.9 30.4
CornerNet $2 \times$ 24.6 49.5 21.7 6.5 20.5 32.2 43.8
CenterNet $70e$ 21.5 48.8 15.6 5.1 16.2 29.6 42.4
Deformable-DETR $50e$ 19.2 44.8 13.7 6.3 15.4 24.9 34.2
Sparse RCNN $1 \times$ 24.2 50.3 20.3 8.8 20.4 30.2 39.4
Faster RCNN $1 \times$ 28.9 59.4 24.1 13.8 25.7 34.5 43.0
Cascade RPN $1 \times$ 29.1 56.5 25.9 12.5 25.5 35.4 44.7
RFLA $1 \times$ 29.7 60.2 25.2 13.2 26.9 35.4 44.6
CFINet $1 \times$ 30.7 60.8 26.7 14.7 27.8 36.4 44.6
Ours $1 \times$ 31.0 60.8 27.1 14.9 27.6 36.9 45.8

Result on SODA-A

Method Schedule $AP$ $AP_{50}$ $AP_{75}$ $AP_{eS}$ $AP_{rS}$ $AP_{gS}$ $AP_N$
Rotated RetinaNet $1 \times$ 26.8 63.4 16.2 9.1 22.0 35.4 28.2
$S^2$ A-Net $1 \times$ 28.3 69.6 13.1 10.2 22.8 35.8 29.5
Oriented RepPoints $1 \times$ 26.3 58.8 19.0 9.4 22.6 32.4 28.5
DHRec $1 \times$ 30.1 68.8 19.8 10.6 24.6 40.3 34.6
Rotated Faster RCNN $1 \times$ 32.5 70.1 24.3 11.9 27.3 42.2 34.4
Gliding Vertex $1 \times$ 31.7 70.8 22.6 11.7 27.0 41.1 33.8
Oriented RCNN $1 \times$ 34.4 70.7 28.6 12.5 28.6 44.5 36.7
DODet $1 \times$ 31.6 68.1 23.4 11.3 26.3 41.0 33.5
CFINet $1 \times$ 34.4 73.1 26.1 13.5 29.3 44.0 35.9
Ours $1 \times$ 34.8 73.6 26.4 13.8 29.7 44.7 36.5

📧 Contact

If you have any problems about this repo or SODA benchmark, please be free to contact us at shaunyuan@mail.nwpu.edu.cn 😉