Rethinking Learnable Tree Filter for Generic Feature Transform

• Main Authors: Song, Lin, Li, Yanwei, Jiang, Zhengkai, Li, Zeming, Zhang, Xiangyu, Sun, Hongbin, Sun, Jian, Zheng, Nanning
• Format: Journal Article
• Language: English
• Published: 07.12.2020
• Subjects: Computer Science - Artificial Intelligence; Computer Science - Computer Vision and Pattern Recognition
• DOI: 10.48550/arxiv.2012.03482

The Learnable Tree Filter presents a remarkable approach to model structure-preserving relations for semantic segmentation. Nevertheless, the intrinsic geometric constraint forces it to focus on the regions with close spatial distance, hindering the effective long-range interactions. To relax the geometric constraint, we give the analysis by reformulating it as a Markov Random Field and introduce a learnable unary term. Besides, we propose a learnable spanning tree algorithm to replace the original non-differentiable one, which further improves the flexibility and robustness. With the above improvements, our method can better capture long-range dependencies and preserve structural details with linear complexity, which is extended to several vision tasks for more generic feature transform. Extensive experiments on object detection/instance segmentation demonstrate the consistent improvements over the original version. For semantic segmentation, we achieve leading performance (82.1% mIoU) on the Cityscapes benchmark without bells-and-whistles. Code is available at https://github.com/StevenGrove/LearnableTreeFilterV2.