Research map / Detection and segmentation / 3D detection and driving
Find n' Propagate: Open-Vocabulary 3D Object Detection in Urban Environments
Djamahl Etchegaray, Zi Helen Huang, Tatsuya Harada, Yadan Luo
Read the paperCheck your idea against 137,217 papers
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 point-dropout density simulator on top of GT sampling (ablation) | Find n' Propagate (Transfusion) · nuScenes Setting 1 | AP_N: 21.14 -> 30.00 | Improved |
| add translation noise only to pasted objects (ablation) | Geometry simulator · nuScenes Setting 1 | AP_N: AP_N 21.14 -> 19.41; AR_N 50.59 -> 56.40 | Mixed |
| add rotation variance on top of translation variance (and dropout) (ablation) | Geometry simulator · nuScenes Setting 1 | AP_N: 19.41 -> 33.65 | Improved |
| combine density ranking with multi-view alignment in the Box Oracle (ablation) | Greedy Box Oracle · nuScenes Setting 3 | AP_N: 15.97 -> 16.72 | Improved |
| add density ranking to alignment-only scoring (ablation) | Greedy Box Oracle · nuScenes Setting 3 | AP_N: 15.05 -> 16.72 | Improved |
| replace FGR rectangular-prior box regression with Greedy Box Seeker | bottom-up proposals · KITTI val pedestrian/cyclist | AP_3D: Cyclist Moderate 3.28 -> 8.02; Pedestrian Moderate 4.51 -> 14.29 | Improved |
| replace OpenSight proposals with Greedy Box Seeker boxes | novel box proposals · nuScenes Setting 3 | AP_N: 5.80 -> 16.72 | Improved |
| add self-training with Box Seeker pseudo-labels instead of top-down self-train | Top-down Self-train (CLIP, Transfusion) · nuScenes Setting 1 | AP_N: 6.47 -> 33.65 | Improved |
| self-training with GLIP labels on top-down proposals | Top-down Projection (GLIP) · nuScenes Setting 1 | AP_N: 4.53 -> 3.79 | Hurt |
| add self-training to CLIP top-down projection | Top-down Projection (CLIP) · nuScenes Setting 1 | AP_N: 3.06 -> 6.47 | Improved |
| use logit fusion with base detector in top-down projection | Top-down Projection (GLIP, Transfusion) · nuScenes Setting 1 | mAP: 24.23 -> 30.64 | Improved |
| swap Transfusion for Centerpoint detector in top-down projection | Top-down Projection · nuScenes Setting 1 | mAP | Hurt |
Closest papers by meaning
- Unlocking Textual and Visual Wisdom: Open-Vocabulary 3D Object Detection Enhanced by Comprehensive Guidance from Text and Image (ECCV 2024 · similarity 0.92)
- OpenSight: A Simple Open-Vocabulary Framework for LiDAR-Based Object Detection (ECCV 2024 · similarity 0.91)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.91)
- Training an Open-Vocabulary Monocular 3D Detection Model without 3D Data (NeurIPS 2024 · similarity 0.90)
- Hierarchical Cross-Modal Alignment for Open-Vocabulary 3D Object Detection (AAAI 2025 · similarity 0.89)
- 3D-AVS: LiDAR-based 3D Auto-Vocabulary Segmentation (CVPR 2025 · similarity 0.89)
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.
