arXiv (Cornell University)· 2017· Preprint
PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric\n Space
- 7,082citations
- 2017year
Short summary
PointNet++ introduces a hierarchical neural network that recursively applies PointNet on nested partitions of point sets, enabling learning of local features at increasing contextual scales by exploiting metric space distances.
AI-generated from the title and abstract; the full text is not read.
Key points
- PointNet++ uses a hierarchical structure to learn features from point sets at multiple scales.
- It exploits metric space distances to capture local structures that PointNet misses.
- Novel set learning layers adapt to varying point densities, enhancing performance.
- The network achieves superior results on 3D point cloud benchmarks.
AI-generated from the title and abstract; the full text is not read.
Abstract
Few prior works study deep learning on point sets. PointNet by Qi et al. is a\npioneer in this direction. However, by design PointNet does not capture local\nstructures induced by the metric space points live in, limiting its ability to\nrecognize fine-grained patterns and generalizability to complex scenes. In this\nwork, we introduce a hierarchical neural network that applies PointNet\nrecursively on a nested partitioning of the input point set. By exploiting\nmetric space distances, our network is able to learn local features with\nincreasing contextual scales. With further observation that point sets are\nusually sampled with varying densities, which results in greatly decreased\nperformance for networks trained on uniform densities, we propose novel set\nlearning layers to adaptively combine features from multiple scales.\nExperiments show that our network called PointNet++ is able to learn deep point\nset features efficiently and robustly. In particular, results significantly\nbetter than state-of-the-art have been obtained on challenging benchmarks of 3D\npoint clouds.\n
The authors' abstract, as published at the source. arXiv (Cornell University), 2017 · DOI ↗
Continue with a free account
Ask the paper: 3 free questions a day about this paper; save it, get its citation, new summaries every day for your field. Takeaways are Premium.
Continue free on the webSign in with Google or Apple; no card needed. You come back to this paper.
On your phone:
Field: Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionComputer Science