SOLOv2: Dynamic and Fast Instance Segmentation
Xinlong Wang1Rufeng Zhang2Tao Kong3Chunhua Shen1Lei Li3
1The University of Adelaide,  2Tongji University,  3ByteDance AI Lab
In this work, we aim at building a simple, direct, and fast instance segmentation framework with strong performance. We follow the principle of the SOLO method. Importantly, we take one step further by dynamically learning the mask head of the object segmenter such that the mask head is conditioned on the location. Specifically, the mask branch is decoupled into a mask kernel branch and mask feature branch, which are responsible for learning the convolution kernel and the convolved features respectively. Moreover, we propose Matrix NMS (non maximum suppression) to significantly reduce the inference time overhead due to NMS of masks. Our Matrix NMS performs NMS with parallel matrix operations in one shot, and yields better results.
SOLOv2 also performs well on object detection and panoptic segmentation.
Demo video here.
Visualization of instance segmentation results using the Res-101-FPN backbone. The model is trained on the COCO train2017 dataset, achieving a mask AP of 39.7 on the COCO test-dev.
  title={SOLOv2: Dynamic and Fast Instance Segmentation},
  author={Wang, Xinlong and Zhang, Rufeng and  Kong, Tao and Li, Lei and Shen, Chunhua},
  journal={Proc. Advances in Neural Information Processing Systems (NeurIPS)},
  title     =  {{SOLO}: Segmenting Objects by Locations},
  author    =  {Wang, Xinlong and Kong, Tao and Shen, Chunhua and Jiang, Yuning and Li, Lei},
  booktitle =  {Proc. Eur. Conf. Computer Vision (ECCV)},
  year      =  {2020}