Research map / Point clouds / Segmentation and detection
Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use instance-level disparity estimation with shape-prior pseudo-ground-truth instead of full-frame disparity | stereo 3D detection on KITTI · KITTI validation, car moderate, IoU=0.7 | AP3d: 27.4 -> 47.15 | Helped |
| use instance-level disparity estimation and shape-derived pseudo-ground-truth | stereo 3D detection on KITTI · KITTI validation, car moderate, IoU=0.7, with LiDAR supervision | APbev: 56.0 -> 64.38 | Helped |
| use Disp R-CNN instead of PL* (AVOD) | stereo 3D detection on KITTI · KITTI test, car moderate, IoU=0.7 | AP3d: 44.6 -> 37.91 | Hurt |
| crop and align object RoIs and estimate disparities only on foreground objects | iDispNet · KITTI validation, foreground pixels, PGT ground truth | pixel-wise disparity EPE: 1.53 -> 0.90 | Helped |
| crop and align object RoIs and estimate disparities only on foreground objects | iDispNet · KITTI validation, foreground pixels, PGT ground truth | object-wise depth RMSE: 1.00 -> 0.33 | Helped |
| reuse RPN features and train the disparity head end-to-end rather than use a separate crop-image encoder-decoder (ablation) | iDispNet · KITTI validation, foreground pixels, PGT ground truth | pixel-wise disparity EPE: 0.90 -> 1.22 | Hurt |
| reuse RPN features and train the disparity head end-to-end rather than use a separate crop-image encoder-decoder (ablation) | iDispNet · KITTI validation, foreground pixels, PGT ground truth | object-wise depth RMSE: 0.33 -> 0.86 | Hurt |
| add dimension regularization during shape optimization (ablation) | shape-based pseudo-ground-truth generation · KITTI object instances | objects with more than 70% vertices inside 3D box: 71% -> 82% | Helped |
| use stereo-network point clouds rather than sparse LiDAR points for pseudo-ground-truth shape optimization | shape-based pseudo-ground-truth generation · KITTI validation, car moderate, IoU=0.7 | AP3d: 37.91 -> 39.34 | No clear effect |
| use Disp R-CNN rather than full-frame PL (AVOD) | KITTI object detection inference · KITTI object detection inference | running time: 0.51 -> 0.425 | Helped |
Closest papers by meaning
- Stereo R-CNN Based 3D Object Detection for Autonomous Driving (CVPR 2019 · similarity 0.90)
- DispViT: Direct Stereo Disparity Regression with a Single-Stream Vision Transformer (ICLR 2026 · similarity 0.89)
- DSGN: Deep Stereo Geometry Network for 3D Object Detection (CVPR 2020 · similarity 0.88)
- Triangulation Learning Network: From Monocular to Stereo 3D Object Detection (CVPR 2019 · similarity 0.88)
- Uni3D: A Unified Baseline for Multi-Dataset 3D Object Detection (CVPR 2023 · similarity 0.88)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.87)
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.