Research map / Detection and segmentation / 3D detection and driving
DSGN++: Exploiting Visual-Spatial Relation for Stereo-Based 3D Detectors
Yilun Chen, Shijia Huang, Shu Liu, Bei Yu, Jiaya Jia
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DSGN++ techniques to DSGN | DSGN · KITTI val (Car) | AP3D: 56.09 -> 61.62 | Improved |
| add DSGN++ techniques to L-DSGN | L-DSGN · KITTI val (Car) | AP3D: 63.58 -> 69.12 | Improved |
| add depth-wise plane sweeping (D-PS) (ablation) | front-view pipeline · KITTI val (Car) | AP3D: 63.58 -> 66.42 | Improved |
| add D-PS to top-view pipeline (ablation) | top-view with FSD head · KITTI val (Car) | AP3D: 61.33 -> 64.59 | Improved |
| add D-PS to monocular top-view pipeline (ablation) | monocular top-view · KITTI val (Car, monocular) | AP3D: 15.36 -> 17.41 | Improved |
| use D-PS instead of Group-PS (ablation) | front-view pipeline · KITTI val (Car) | AP3D: 64.98 -> 66.42 | Improved |
| add front-surface depth head instead of voxel occupancy (ablation) | top-view pipeline · KITTI val (Car) | AP3D: 56.69 -> 61.33; 61.46 -> 64.43 | Improved |
| integrate front-view and top-view volumes (dual-view) (ablation) | DSGN++ · KITTI val (Car) | AP3D: 67.63 -> 69.12; 67.85 -> 69.12 | Improved |
| add Stereo-LiDAR copy-paste (ablation) | dual-view DSGN++ · KITTI val (Car/Ped./Cyc.) | AP3D: 66.21 -> 69.12; 37.19 -> 42.44; 30.01 -> 42.48 | Improved |
| add occluded point removal in copy-paste (ablation) | SLCP · KITTI val (Car/Ped./Cyc.) | AP3D: 66.52 -> 69.12; 38.79 -> 42.44; 36.27 -> 42.48 | Improved |
| raise copy-paste probability from 0.6 to 1. (ablation) | SLCP · KITTI val (Car) | AP3D: 69.12 -> 68.55 | Hurt |
| increase pasted samples from 5 to 7 per class (ablation) | SLCP · KITTI val (Car) | AP3D: 69.12 -> 68.92 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DSGN — DSGN: Deep Stereo Geometry Network for 3D Object Detection (CVPR 2020) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DSGN: Deep Stereo Geometry Network for 3D Object Detection (CVPR 2020 · similarity 0.92)
- LIGA-Stereo: Learning LiDAR Geometry Aware Representations for Stereo-Based 3D Detector (ICCV 2021 · similarity 0.88)
- Advancing 3D Object Detection With Depth-Aware Spatial Knowledge Distillation (TPAMI 2025 · similarity 0.87)
- RayD3D: Distilling Depth Knowledge Along the Ray for Robust Multi-View 3D Object Detection (AAAI 2026 · similarity 0.87)
- RefineAny3D: Depth Refinement as Semantic Alignment for Monocular 3D Detection (NeurIPS 2026 · similarity 0.86)
- Towards Generalizable Multi-Camera 3D Object Detection via Perspective Rendering (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
