Research map / Detection and segmentation / 3D detection and driving
Probing the 3D Object-Level Understanding of Pre-Trained Detection Transformers
Robin Kim, Colin Samplawski, Benjamin Marlin
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 |
|---|---|---|---|
| probe query embedding instead of 2D bounding box (linear) (ablation) | DETR · Virtual KITTI 2 depth | MAE (m): 4.27 -> 1.01 | Improved |
| probe query embedding instead of 2D bounding box (MLP) (ablation) | DETR · Virtual KITTI 2 depth | MAE (m): 1.02 -> 0.54 | Improved |
| probe query embedding instead of 2D bounding box (linear) (ablation) | DETR · NYUv2 depth | MAE (m): 0.87 -> 0.58 | Improved |
| probe query embedding instead of 2D bounding box (MLP) (ablation) | DETR · NYUv2 depth | MAE (m): 0.77 -> 0.51 | Improved |
| use MLP probe instead of linear probe (ablation) | DETR · Virtual KITTI 2 depth | MAE (m): 1.01 -> 0.54 | Improved |
| use MLP probe instead of linear probe (ablation) | DETR · NYUv2 depth | MAE (m): 0.58 -> 0.51 | Improved |
| use MLP probe instead of linear probe (ablation) | RT-DETR v2 · Virtual KITTI 2 depth | MAE (m): 1.56 -> 1.05 | Improved |
| learn an MLP output head on Depth Anything 2 instead of zero-shot (ablation) | Depth Anything 2 · Virtual KITTI 2 depth | MAE (m): 1.49 -> 0.50 | Improved |
| learn an MLP output head on Depth Anything 2 instead of zero-shot (ablation) | Depth Anything 2 · NYUv2 depth | MAE (m): 0.61 -> 0.36 | Improved |
| learn an MLP output head on MonoDETR instead of zero-shot (ablation) | MonoDETR · NYUv2 depth | MAE (m): 10.59 -> 0.75 | Improved |
| compare real-time detectors RT-DETR v2 against DETR with MLP probe | RT-DETR v2 · Virtual KITTI 2 depth | MAE (m): 0.54 -> 1.05 | Hurt |
| compare 3D-supervised MonoDETR against 2D DETR MLP probe | DETR · KITTI 3D center prediction | Center MAE (m): 0.50 vs 0.30 | Hurt |
Closest papers by meaning
- End-to-End Object Detection with Transformers (ECCV 2020 · similarity 0.90)
- Efficient Decoder-Free Object Detection with Transformers (ECCV 2022 · similarity 0.89)
- An End-to-End Transformer Model for 3D Object Detection (ICCV 2021 · similarity 0.89)
- OA-DET3D: Embedding Object Awareness As A General Plug-in for Multi-Camera 3D Object Detection (IJCV 2025 · similarity 0.88)
- Detect Anything 3D in the Wild (ICCV 2025 · similarity 0.88)
- V-DETR: DETR with Vertex Relative Position Encoding for 3D Object Detection (ICLR 2024 · 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.
