2019/05/17 by Zhizhong Han, Xinhai Liu, Han, Zhizhong +5
Computer Science · Earth and Planetary Sciences · Engineering · #3D Shape Modeling and Analysis #3D Surveying and Cultural Heritage #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences
paper · pdf · doi:10.48550/arxiv.1905.07506
openalex publication_date 2019/05/17 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Deep learning has achieved remarkable results in 3D shape analysis by\nlearning global shape features from the pixel-level over multiple views.\nPrevious methods, however, compute low-level features for entire views without\nconsidering part-level information. In contrast, we propose a deep neural\nnetwork, called Parts4Feature, to learn 3D global features from part-level\ninformation in multiple views. We introduce a novel definition of generally\nsemantic parts, which Parts4Feature learns to detect in multiple views from\ndifferent 3D shape segmentation benchmarks. A key idea of our architecture is\nthat it transfers the ability to detect semantically meaningful parts in\nmultiple views to learn 3D global features. Parts4Feature achieves this by\ncombining a local part detection branch and a global feature learning branch\nwith a shared region proposal module. The global feature learning branch\naggregates the detected parts in terms of learned part patterns with a novel\nmulti-attention mechanism, while the region proposal module enables locally and\nglobally discriminative information to be promoted by each other. We\ndemonstrate that Parts4Feature outperforms the state-of-the-art under three\nlarge-scale 3D shape benchmarks.\n