NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:964
Title:Learnable Tree Filter for Structure-preserving Feature Transform

The paper introduces learnable tree filter using minimal spanning tree for modeling long-range dependencies. The proposed algorithm is linear-time and can be incorporated into commonly used deep neural network. Empirical evaluation shows leading performance with ResNet-101 on PASCAL VOC 2012. All reviewers found the contributions of this work significant, both from methodological and empirical perspectives, and rated the paper positively. I recommend accept for this paper.