Where LiDAR meets literature: a 3D detection reading list

What it does

This repository is a curated bibliography of deep-learning research focused on 3D object detection from LiDAR point clouds. It collects and links scores of papers, points to official code repositories where available, and catalogs standard autonomous-driving datasets including KITTI, nuScenes, Waymo, and Lyft with training and validation sample counts.

The interesting bit

Rather than trying to be a broad computer-vision wiki, the list stays narrowly focused on LiDAR-based methods, tagging many entries with their conference origin — CVPR, ICCV, ECCV — and linking to related autonomous-driving workshops. That specificity makes it a decent snapshot of the field’s trajectory from early voxel-based detectors to recent transformer-fusion approaches.

Key highlights

• Explicitly focused on LiDAR-based methods; the title itself draws that boundary
• Dataset cheat-sheet with exact sample counts for KITTI (3,712 train / 3,769 val), nuScenes (28k train / 6k val), Waymo (798 sequences, ~158k LiDAR samples), and Lyft
• Papers range from foundational work (PointNet, VoxelNet) through CVPR 2022 entries (TransFusion, SST, PDV)
• Many entries include direct arXiv or PDF links; some also link to official code repositories
• Includes workshop tracks and challenges (nuScenes, Waymo, BDD1k) tied to specific conference years

Caveats

• This is pure curation — no installable code, no benchmark framework, no unified API
• The entire index lives in a single README, so navigation is basically Ctrl+F
• A minority of paper links route through Baidu, Zhihu, or onacademic rather than stable preprint servers, so rot is a risk

Verdict

Worth bookmarking if you’re doing a literature review or entering the autonomous-driving perception space. Skip it if you need a ready-to-run training pipeline or comparative benchmark suite.