Grapy-ML: Graph Pyramid Mutual Learning for Cross-Dataset Human Parsing

• DOI: 10.1609/aaai.v34i07.6728
• Publication Date: 2020-06-29T18:43:13Z
• Authors: Haoyu He; Jing Zhang; Qiming Zhang; Dacheng Tao
• Keywords: Granularity; Discriminative model; Pyramid (geometry); Feature Learning; Feature (linguistics)

Human parsing, or human body part semantic segmentation, has been an active research topic due to its wide potential applications. In this paper, we propose a novel GRAph PYramid Mutual Learning (Grapy-ML) method address the cross-dataset parsing problem, where annotations are at different granularities. Starting from prior knowledge of hierarchical structure, devise graph pyramid module (GPM) by stacking three levels structures coarse granularity fine subsequently. At each level, GPM utilizes self-attention mechanism model correlations between context nodes. Then, it adopts top-down progressively refine features through all levels. also enables efficient mutual learning. Specifically, network weights first two shared exchange learned coarse-granularity information across datasets. By making use multi-granularity labels, Grapy-ML learns more discriminative feature representation and achieves state-of-the-art performance, which is demonstrated extensive experiments on popular benchmarks, e.g. CIHP dataset. The source code publicly available https://github.com/Charleshhy/Grapy-ML.