The official pytorch implementation of CPM R-CNN.

CPM R-CNN: Calibrating Point-guided Misalignment in Object Detection,
Bin Zhu, Qing Song, Lu Yang, Zhihui Wang, Chun Liu, Mengjie Hu WACV 2021. (arXiv pre-print)

In object detection, offset-guided and point-guided regression dominate anchor-based and anchor-free method separately. Recently, point-guided approach is introduced to anchor-based method. However, we observe points predicted by this way are misaligned with matched region of proposals and score of localization, causing a notable gap in performance. In this paper, we propose CPM R-CNN which contains three efficient modules to optimize anchor-based point-guided method. According to sufficient evaluations on the COCO dataset, CPM R-CNN is demonstrated efficient to improve the localization accuracy by calibrating mentioned misalignment. Compared with Faster R-CNN and Grid R-CNN based on ResNet-101 with FPN, our approach can substantially improve detection mAP by 3.3% and 1.5% respectively without whistles and bells. Moreover, our best model achieves improvement by a large margin to 49.9% on COCO test-dev. Code and models will be publicly available.

• CPM R-CNN pipeline:

• Modules in CPM R-CNN:

• 8 x TITAN RTX GPU

• pytorch1.5.1

• python3.7.0

• Other details will be public soon.

On MS COCO test-dev

Component-wise performance

ImageNet pretrained weight

• R-50
• R-101
• X-101-64x4d

To train a model with 8 GPUs run:

python -m torch.distributed.launch --nproc_per_node=8 tools/rcnn/train_net.py --cfg $CFG_NAME

python tools/rcnn/test_net.py --cfg $CFG_NAME

If you find this work or code is helpful in your research, please cite:

@article{zhu2020cpm,
  title={CPM R-CNN: Calibrating Point-guided Misalignment in Object Detection},
  author={Zhu, Bin and Song, Qing and Yang, Lu and Wang, Zhihui and Liu, Chun and Hu, Mengjie},
  journal={arXiv preprint arXiv:2003.03570},
  year={2020}
}