Research map / 3D vision / 3D generation
VoxFormer: Sparse Voxel Transformer for Camera-Based 3D Semantic Scene Completion
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 |
|---|---|---|---|
| temporal input (current + previous 4 frames, VoxFormer-T) vs single image (VoxFormer-S) | VoxFormer · SemanticKITTI val | mIoU: mIoU +8.10% (51.2m), +11.77% (25.6m), +22.03% (12.8m) | Helped |
| depth-based query proposal (occupancy-filtered) vs dense queries in stage-2 (ablation) | VoxFormer · SemanticKITTI val | IoU / memory: IoU 44.0 vs 34.6 dense; memory 14.6G vs 18.5G | Helped |
| stereo depth instead of monocular depth (ablation) | VoxFormer-S · SemanticKITTI val 51.2m | IoU (%): 44.02 vs 38.68 (mono) | Helped |
| depth estimation removed (no image depth for query proposal) (ablation) | VoxFormer · SemanticKITTI val 51.2m | IoU / mIoU: IoU 44.02 -> 34.64; mIoU 12.35 -> 10.14 | Hurt |
| depth correction (occupancy network) removed (ablation) | VoxFormer · SemanticKITTI val 51.2m | IoU / mIoU: IoU 44.02 -> 36.95; mIoU 12.35 -> 11.36 | Hurt |
| cross-attention to image features removed (ablation) | VoxFormer · SemanticKITTI val 51.2m | IoU / mIoU: IoU 44.02 -> 32.74; mIoU 12.35 -> 9.94 | Hurt |
| self-attention removed from stage-2 (ablation) | VoxFormer · SemanticKITTI val 51.2m | IoU / mIoU: IoU 44.02 -> 43.73 (w/o); mIoU 12.35 -> 10.70 (w/o) | Helped |
| offline future frames added to temporal input (online to full offline window) (ablation) | VoxFormer-T · SemanticKITTI val | mIoU: mIoU 13.24 (online) -> 16.20 (offline, all frames); IoU 44.31 -> 45.05 | Helped |
| VoxFormer-S vs MonoScene camera-based SSC | VoxFormer-S · SemanticKITTI val 51.2m | IoU (%): 44.02 vs 36.80 | Helped |
| random query subset instead of occupancy-based query proposal (ablation) | VoxFormer · SemanticKITTI val | IoU / mIoU: random 10% IoU 24.6 / mIoU 3.8 vs occupancy 44.0 / 12.4; random ratios unstable | Hurt |
Closest papers by meaning
- Context and Geometry Aware Voxel Transformer for Semantic Scene Completion (NeurIPS 2024 · similarity 0.93)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.92)
- Not All Voxels Are Equal: Hardness-Aware Semantic Scene Completion with Self-Distillation (CVPR 2024 · similarity 0.90)
- Voxel Proposal Network via Multi-Frame Knowledge Distillation for Semantic Scene Completion (NeurIPS 2024 · similarity 0.90)
- IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals (NeurIPS 2025 · similarity 0.90)
- Memory-Augmented Re-Completion for 3D Semantic Scene Completion (AAAI 2025 · similarity 0.90)
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.