Research map / 3D vision / 3D generation
LightOcc: Lightweight Spatial Embedding for Efficient Vision-based 3D Occupancy Prediction
Jinqing Zhang, Yanan Zhang, Wenrui Cai, Qingjie Liu, Yunhong Wang
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 Spatial-to-Channel embedding (ablation) | FlashOcc-opt · Occ3D-nuScenes (R50, 256x704) | mIoU: 35.00 -> 36.30 | Improved |
| add Lightweight Cross-View Interaction (2 layers) (ablation) | LightOcc-T · Occ3D-nuScenes (R50, 256x704) | mIoU: 36.30 -> 36.86 | Improved |
| add edge-aware spatial embedding (ablation) | LightOcc-S · Occ3D-nuScenes (R50, 256x704) | mIoU: 38.17 -> 38.46 | Improved |
| add geometric supervision on embeddings (ablation) | LightOcc-S · Occ3D-nuScenes (R50, 256x704) | mIoU: 38.46 -> 38.70 | Improved |
| add BEV-CutMix with more epochs (ablation) | LightOcc-S · Occ3D-nuScenes (R50, 256x704) | mIoU: 38.70 -> 39.24 | Improved |
| add BEV-CutMix with more epochs (ablation) | FlashOcc-opt · Occ3D-nuScenes (R50, 256x704) | mIoU: 35.00 -> 35.89 | Improved |
| use Sigmoid instead of Softmax for depth distribution (ablation) | LightOcc-T · Occ3D-nuScenes (R50, 256x704) | mIoU: 36.11 -> 36.71 | Improved |
| divide multiplied embeddings by vanished dimension (Mean) (ablation) | LightOcc-T · Occ3D-nuScenes (R50, 256x704) | mIoU: 36.71 -> 36.86 | Improved |
| use LCVI instead of TPVFormer cross-view hybrid attention (ablation) | LightOcc-T · Occ3D-nuScenes (R50, 256x704) | latency: 45.63 ms -> 35.23 ms | Improved |
| use two convolutions per interaction layer (ablation) | LightOcc-T · Occ3D-nuScenes (R50, 256x704) | mIoU: 36.86 -> 36.92 | No clear effect |
| use gradient edge instead of Sobel 3D convolution (ablation) | LightOcc-S · Occ3D-nuScenes (R50, 256x704) | latency: 42.57 ms -> 40.38 ms | Improved |
| cut occupancy randomly rather than from center (ablation) | BEV-CutMix · Occ3D-nuScenes (R50, 256x704) | mIoU: 37.93 -> 37.59 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- GaussianWorld — GaussianWorld: Gaussian World Model for Streaming 3D Occupancy Prediction (CVPR 2025) · better on 2 of 2 shared results
- OccGen — OccGen: Generative Multi-modal 3D Occupancy Prediction for Autonomous Driving (ECCV 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LowRankOcc: Tensor Decomposition and Low-Rank Recovery for Vision-based 3D Semantic Occupancy Prediction (CVPR 2024 · similarity 0.89)
- OctOcc: High-Resolution 3D Occupancy Prediction with Octree (AAAI 2024 · similarity 0.89)
- 3D Occupancy Prediction with Low-Resolution Queries via Prototype-aware View Transformation (CVPR 2025 · similarity 0.89)
- AdaOcc: Adaptive 3D Occupancy Prediction for Embodied Tasks (NeurIPS 2026 · similarity 0.89)
- Spatiotemporal Decoupling for Efficient Vision-Based Occupancy Forecasting (CVPR 2025 · similarity 0.89)
- SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction (CVPR 2024 · 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.
