Research map / 3D vision / 3D generation
Learning Ego 3D Representation As Ray Tracing
Jiachen Lu, Zheyuan Zhou, Xiatian Zhu, Hang Xu, Li Zhang
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 adaptive attention (learnable sampling offsets) (ablation) | Ego3RT · nuScenes val | mAP / NDS: mAP 0.365 -> 0.375, NDS 0.441 -> 0.450 | Improved |
| add polarized grid of eyes (vs rectangular grid) (ablation) | Ego3RT without polar attention · nuScenes val | mAP / NDS / mATE: mAP 0.365 -> 0.362, NDS 0.437 -> 0.440, mATE 0.673 -> 0.656 | Mixed |
| add polar attention on eyes of each polar ray (ablation) | Ego3RT with polarized grid · nuScenes val | mAP / NDS: mAP 0.362 -> 0.375, NDS 0.440 -> 0.450 | Improved |
| add 3D feature augmentation (ablation) | Ego3RT baseline · nuScenes val | mAP: mAP 0.345 -> 0.353, NDS 0.421 -> 0.427 | Improved |
| increase N_point from 1 to 3 (look around) (ablation) | Ego3RT · nuScenes val | mAP: mAP 0.353 -> 0.360, NDS 0.427 -> 0.433 | Improved |
| add adaptive attention (component table) (ablation) | Ego3RT · nuScenes val | mAP: mAP 0.360 -> 0.365, NDS 0.433 -> 0.437 | Improved |
| add polarization (grid and polar attention) (ablation) | Ego3RT · nuScenes val | mAP: mAP 0.365 -> 0.375, NDS 0.437 -> 0.450 | Improved |
| use ResNet101 instead of ResNet50 backbone (ablation) | Ego3RT · nuScenes val | mAP / NDS: mAP 0.332 -> 0.355, NDS 0.380 -> 0.409 | Improved |
| initialize from FCOS3D checkpoint (ablation) | Ego3RT ResNet101 · nuScenes val | mAP / NDS: mAP 0.355 -> 0.375, NDS 0.409 -> 0.450 | Improved |
| initialize from DD3D checkpoint with VoVNet (ablation) | Ego3RT · nuScenes val | mAP / NDS: mAP 0.375 -> 0.478, NDS 0.450 -> 0.534 | Improved |
| vary eyes height above ground (0m, 0.4m, 0.8m) (ablation) | Ego3RT · nuScenes val | NDS: NDS 0.445 / 0.437 / 0.450, mAP 0.375 for all | No clear effect |
| increase eyes density from 64x256 to 80x256 (ablation) | Ego3RT · nuScenes val | mAP: mAP 0.366 -> 0.375, 96x256 gives 0.372 | Improved |
Closest papers by meaning
- Parametric Depth Based Feature Representation Learning for Object Detection and Segmentation in Bird's-Eye View (ICCV 2023 · similarity 0.88)
- Learning 3D Perception from Others' Predictions (ICLR 2025 · similarity 0.88)
- Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene (CVPR 2025 · similarity 0.87)
- Learning Ego-Centric BEV Representations from a Perspective-Privileged View: Cross-View Supervision for Online HD Map Construction (ECCV 2026 · similarity 0.87)
- Instance-Aware Multi-Camera 3D Object Detection with Structural Priors Mining and Self-Boosting Learning (AAAI 2024 · similarity 0.87)
- Achieving Speed-Accuracy Balance in Vision-based 3D Occupancy Prediction via Geometric-Semantic Disentanglement (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.
