A field guide to deep learning on messy 3D points

What it does

This is the companion repository to a 2020 IEEE TPAMI survey on deep learning for 3D point clouds. It does not ship code or models; instead, it serves as a curated reference, organizing the field into a visual taxonomy and collecting benchmark results across classification, detection, and segmentation tasks. The authors also provide a consolidated index of public datasets and their download portals.

The interesting bit

The repository treats a dense academic literature review as a living lookup tool. Rather than burying insights in PDF tables, it surfaces taxonomy diagrams and leaderboard-style result images that let you scan the state of the field in seconds.

Key highlights

• Visual taxonomy diagram categorizing deep learning methods for point cloud analysis
• Benchmark result tables (as images) for 3D shape classification, object detection, and segmentation
• Curated links to major datasets including KITTI, SemanticKITTI, ModelNet, S3DIS, ScanNet, and nuScenes
• Official companion to a peer-reviewed IEEE TPAMI paper with an open arXiv preprint
• Accepts community updates via issues and pull requests

Caveats

• The update log shows no activity since February 2020, so newer methods published after that date are unlikely to appear
• This is a reference page, not a framework; there are no scripts, models, or training pipelines to run
• A few README typos (e.g., “objecct”) suggest the page has not seen recent polish

Verdict

Newcomers to 3D vision and researchers hunting for dataset links or baseline comparisons should start here. If you need a training framework or model zoo, this is not it.