1st Place in COCO 2017 Instance Segmentation, 2nd Place in COCO 2017 Object Detection, Outperforms Mask R-CNN, FCIS, G-RMI, & RetinaNet
In this story, Path Aggregation Network (PANet), by The Chinese University of Hong Kong, Peking University, SenseTime Research, and YouTu Lab, Tencent, is presented. In PANet:
- Bottom-up path augmentation is used to shorten the information path between lower layers and topmost feature.
- Adaptive feature pooling is used to link feature grids at all feature levels.
- Fully connected fusion is used to improve the mask prediction.
- Finally, it win the 1st place in COCO 2017 Challenge Instance Segmentation task, and 2nd place in Object Detection task without large-batch training.
This is a paper in 2018 CVPR with over 300 citations. (Sik-Ho Tsang @ Medium)