Research map / Detection and segmentation / 3D detection and driving
ConQueR: Query Contrast Voxel-DETR for 3D Object Detection
Read the paperCheck 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 |
|---|---|---|---|
| Add Query Contrast to Voxel-DETR | Voxel-DETR · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: 1.1, 1.7, 2.3 | Helped |
| Add Query Contrast to Voxel-DETR | Voxel-DETR · Waymo Open Dataset validation | mAPH/L2: over 1.6 | Helped |
| Use score thresholding instead of fixed top-N predictions | ConQueR · Waymo Open Dataset validation | predictions per scene: 300 -> 131 | Helped |
| Remove multi-positive noised GT copies (ablation) | ConQueR · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: -0.8, -0.6, -0.5 | Hurt |
| Remove auxiliary denoising loss (ablation) | ConQueR without multi-positive pairs · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: +0.1, +0.1, -0.3 | Mixed |
| Remove InfoNCE Query Contrast (ablation) | ConQueR · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: -0.4, -1.2, -1.5 | Hurt |
| Replace Query Contrast with positive-pair-only KD-MSE (ablation) | ConQueR · Waymo Open Dataset validation, especially pedestrian and cyclist | APH/L2: 63.4 vs 64.7; 68.2 vs 70.1 | Hurt |
| Replace cosine similarity with GIoU similarity (ablation) | ConQueR · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: 66.6 vs 68.2; 63.6 vs 64.7; 68.4 vs 70.1 | Mixed |
| Use 3 noised GT groups without original GTs rather than include original GTs (ablation) | ConQueR · Waymo Open Dataset validation | APH/L2 pedestrian: 64.3 -> 64.7 | Helped |
| Use 4 rather than 3 noised GT groups (ablation) | ConQueR · Waymo Open Dataset validation | APH/L2 vehicle: 68.2 -> 67.7 | Hurt |
| Remove query-to-GT feature projection (ablation) | ConQueR · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: 67.2 vs 68.2; 64.2 vs 64.7; 69.3 vs 70.1 | Hurt |
| Apply projection to both GT and query embeddings (ablation) | ConQueR · Waymo Open Dataset validation, vehicle/pedestrian/cyclist | APH/L2: 67.3 vs 68.2; 64.1 vs 64.7; 68.9 vs 70.1 | Hurt |
Closest papers by meaning
- Redundant Queries in DETR-Based 3D Detection Methods: Unnecessary and Prunable (AAAI 2026 · similarity 0.89)
- DiffuDETR: Rethinking Detection Transformers with Diffusion Process (ICLR 2026 · similarity 0.88)
- V-DETR: DETR with Vertex Relative Position Encoding for 3D Object Detection (ICLR 2024 · similarity 0.88)
- Far3D: Expanding the Horizon for Surround-View 3D Object Detection (AAAI 2024 · similarity 0.87)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.86)
- GO-N3RDet: Geometry Optimized NeRF-enhanced 3D Object Detector (CVPR 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.