PointOBB-v2

This is the repo of PointOBB-v2: Towards Simpler, Faster, and Stronger Single Point Supervised Oriented Object Detection

Installation

Please refer to the Installation, we copy it here.

conda create -n open-mmlab python=3.7 pytorch==1.7.0 cudatoolkit=10.1 torchvision -c pytorch -y
conda activate open-mmlab
pip install openmim
mim install mmcv-full
mim install mmdet
git clone https://github.com/taugeren/PointOBB-v2.git
cd mmrotate
pip install -r requirements/build.txt
pip install -v -e .

Data Preparation

Please follow data_preparation to prepare formatting data

Train CPM

If you want to visualize CPM result during training, please set visualize=True in train_config

Please modified the config code that contains the visualize directory path

The learning rate for n GPU card and batch size m is 0.0125 * n * m

For single GPU

# Basic format: python tools/train.py ${CONFIG_FILE} [optional arguments]
python tools/train.py configs/pointobbv2/train_cpm_dotav10.py --work-dir work_dirs/cpm_dotav10 --gpu-ids 0

For multiple GPU

# Basic format: ./tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM} [optional arguments]
CUDA_VISIBLE_DEVICES=0,1 PORT=29801 ./tools/dist_train.sh configs/pointobbv2/train_cpm_dotav10.py 2

Generate Pseudo Label

Please modified the config code that contains the directory path

For single GPU

# Basic fromat: python tools/train.py ${CONFIG_FILE} --resume-from ${CPM_CHECKPOINT_FILE} [other arguments]
python tools/train.py configs/pointobbv2/generate_pseudo_label_dotav10.py --resume-from work_dirs/cpm_dotav10/epoch_6.pth --work-dir work_dirs/cpm_dotav10 --gpu-ids 0

For multiple GPU

# Basic format: ./tools/dist_train_resume.sh ${CONFIG_FILE} ${CPM_CHECKPOINT_FILE} ${GPU_NUM} [optional arguments]
CUDA_VISIBLE_DEVICES=0,1 PORT=29801 ./tools/dist_train_resume.sh /ssd1/renbotao/github_submission/mmrotate/configs/pointobbv2/generate_pseudo_label_dotav10.py work_dirs/cpm_dotav10/epoch_6.pth 2

Train Detector

You can use different oriented object detection detector in MMRotate Config

Please modify the pseudo label path in config file

For example, using Redet, the training command:

# single GPU
python tools/train.py configs/pointobbv2/redet_dotav10.py --work-dir work_dirs/cpm_dotav10 --gpu-ids 0

# multiple GPU
CUDA_VISIBLE_DEVICES=0,1 PORT=29801 ./tools/dist_train.sh configs/pointobbv2/redet_dotav10.py 2

the testing command:

# single GPU
python tools/test.py work_dirs/redet_dotav10/redet_dotav10.py work_dirs/redet_dotav10/epoch_12.pth --gpu-ids 0 --format-only --eval-options submission_dir=testmodel/redet_dotav10_epoch12

# multiple GPU
CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29816 tools/dist_test1.sh work_dirs/redet_dotav10/redet_dotav10.py work_dirs/redet_dotav10/epoch_12.pth 4

Our log and checkpoint