object_detection

Applying SPANet to object detection and instance segmentation

Environement Setup

Install MMDetection v2.28.1.

• Check mmcv-full version depending on your pytorch version.

# Compatible with PyTorch 1.12.1 + CUDA 11.6
pip install timm==0.6.12
pip install -U openmim 
mim install mmcv-full==1.6.0

python setup.py develop

Data preparation

Prepare COCO 2017 according to the guidelines in MMDetection. If you have interest in this dataset, refer to below links:

• mmdetection docs
• COCO homepage

Evaluation

To evaluate SPANet-S24 + RetinaNet on a single node with 4 GPUs run:

#  evaluation by default
FORK_LAST3=1 ./tools/dist_test.sh local_configs/fpn_spanet/S24/retinanet_spanet_s24_fpn_1x_coco.py /path/to/checkpoint_file 4 --out results.pkl --eval bbox

# simple usage 
bash run_eval.sh

To evaluate SPANet-S24 + RetinaNet on a single node with 4 GPUs run:

#  evaluation by default
./tools/dist_test.sh local_configs/fpn_spanet/S24/mask_rcnn_spanet_s24_fpn_1x_coco.py /path/to/checkpoint_file 4 --out results.pkl --eval bbox segm

# simple usage 
bash run_eval.sh

Training

To train SPANet-S24 + RetinaNet on a single node with 4 GPUs run:

#  training by default
FORK_LAST3=1 ./tools/dist_train.sh local_configs/fpn_spanet/S24/retinanet_spanet_s24_fpn_1x_coco.py 4

# simple usage 
bash run_train.sh

To train SPANet-S24 + Mask R-CNN on a single node with 4 GPUs run:

#  training by default
./tools/dist_train.sh local_configs/fpn_spanet/S24/mask_rcnn_spanet_s24_fpn_1x_coco.py 4

# simple usage 
bash run_train.sh

MACs

To calculate MACs for a model, run:

python tools/analysis_tools/get_flops.py /path/to/config

# simple usage
bash get_flops.sh

Bibtex

@article{yun2023spanet,
  title={SPANet},
  author={Yun, Guhnoo},
  journal={arXiv preprint arXiv:xxxx.xxxxx},
  year={2023}
}

Acknowledgment

Our implementation is mainly based on the following codebases. We gratefully thank the authors for their wonderful works.

mmdetection, LITv2, PoolFormer.