Skip to content

Latest commit

 

History

History

mva2023_sod4bird

README.md

MVA2023 SOD4SB

MVA2023 Small Object Detection Challenge for Spotting Birds

Results and Models

Method Backbone Lr schd Scale Flip val AP val AP50 public test AP50 Config Download
Faster R-CNN R-50 7e 2160 N 36.2 68.1 54.8 config model
GFL R-50 7e 2160 N 46.5 88.6 72.4 config model
GFL R-50 7e 2160 Y 47.0 88.9 73.1 config model
  • Scale: shorter side pixels.
  • Flip: horizontal flip for test-time augmentation.

Usage

Installation

See get_started.md to install Miniconda and CUDA toolkit.

# change channel setting of conda
conda config --remove channels defaults
conda config --add channels conda-forge
# create and activate conda environment
conda create --name univmva2023 python=3.9 -y
conda activate univmva2023
# install major requirements
conda install pytorch==1.9.0 torchvision==0.10.0 cudatoolkit=11.1 -c pytorch -c conda-forge
pip install mmcv-full==1.4.4
# install UniverseNet
git clone https://github.com/shinya7y/UniverseNet.git
cd UniverseNet
pip install -v -e .

My testing environment is as follows. Version details can be found in this conda environment file.

  • GPU: NVIDIA GeForce RTX 3090
  • Ubuntu 22.04 (on WSL2)
  • Python 3.9
  • PyTorch 1.9.0
  • cudatoolkit 11.1
  • mmcv-full 1.4.4
  • nvcc 11.7
  • GCC 11.3

If you have any problems with the above installation, apt install might be required beforehand. For example, sudo apt -y install build-essential libgl1-mesa-glx and other libraries written in Dockerfile.

To resolve installation problems, please see the documents and issues of MMDetection and MMCV.

Preparation

Based on your data directory and evaluation dataset (val / public test / private test), please edit data_root, data.test.ann_file, and data.test.img_prefix in the configs in this directory.

Testing

# download trained model
mkdir -p ${HOME}/data/checkpoints/
wget -P ${HOME}/data/checkpoints/ https://github.com/shinya7y/weights/releases/download/v1.0.3/gfl_r50_fpn_fp16_1x2x8_7e_mva2023_sod4bird_2160_20230415_epoch_7-383174fb.pth

CONFIG_FILE=configs/datasets/mva2023_sod4bird/gfl_r50_fpn_fp16_1x2x8_7e_mva2023_sod4bird_2160_test.py
CHECKPOINT_FILE=${HOME}/data/checkpoints/gfl_r50_fpn_fp16_1x2x8_7e_mva2023_sod4bird_2160_20230415_epoch_7-383174fb.pth
# for evaluation
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --eval bbox
# for submission
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --format-only --eval-options jsonfile_prefix=results

To help reproduction with other repositories that are based on mmdet (around 2.25.0), I provide another config. Note that this config is only for reproduction of testing.

Training

CONFIG_FILE=configs/datasets/mva2023_sod4bird/gfl_r50_fpn_fp16_1x2x8_7e_mva2023_sod4bird_2160.py
CONFIG_NAME=$(basename ${CONFIG_FILE} .py)
WORK_DIR="${HOME}/logs/mva2023_sod4bird/${CONFIG_NAME}_`date +%Y%m%d_%H%M%S`"
python tools/train.py ${CONFIG_FILE} --work-dir ${WORK_DIR} --seed 0

This training takes 5 hours with one RTX 3090.

Used pre-trained model

I used a COCO pre-trained model, which is publicly available from mmdetection/configs/gfl.

Citation

Please check the latest bibliographic information. https://www.mva-org.jp/mva2023/challenge

Dateset:

@inproceedings{SOD4SB_MVA2023,
  title     = {{MVA2023 Small Object Detection Challenge for Spotting Birds: Dataset, Methods, and Results}},
  author    = {Yuki Kondo and Norimichi Ukita and Takayuki Yamaguchi and Hao-Yu Hou and Mu-Yi Shen and Chia-Chi Hsu and En-Ming Huang and Yu-Chen Huang and Yu-Cheng Xia and Chien-Yao Wang and Chun-Yi Lee and Da Huo and Marc A. Kastner and Tingwei Liu and Yasutomo Kawanishi and Takatsugu Hirayama and Takahiro Komamizu and Ichiro Ide and Yosuke Shinya and Xinyao Liu and Guang Liang and Syusuke Yasui},
  booktitle = {International Conference on Machine Vision and Applications (MVA)},
  year      = {2023}
}

Results and models:

@inproceedings{BandRe_MVA2023,
  author    = {Shinya, Yosuke},
  title     = {{BandRe}: Rethinking Band-Pass Filters for Scale-Wise Object Detection Evaluation},
  booktitle = {International Conference on Machine Vision and Applications (MVA)},
  year      = {2023}
}