Research map / 3D vision / 3D generation
CVSformer: Cross-View Synthesis Transformer for Semantic Scene Completion
Haotian Dong, Enhui Ma, Lubo Wang, Miaohui Wang, Wuyuan Xie, Qing Guo, Ping Li, Lingyu Liang, Kairui Yang, Di Lin
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 Multi-View Feature Synthesis (rotated 3D kernels) to the baseline 3D CNN (ablation) | CVSformer · NYU test | SSC mIoU (%): 51.3 -> 51.8 | Improved |
| add Cross-View Transformer on top of MVFS (ablation) | CVSformer · NYU test | SSC mIoU (%): 51.8 -> 52.6 | Improved |
| learn view tokens with convolution plus self-attention instead of either alone (ablation) | CVTr · NYU test | SSC mIoU (%): 52.0 -> 52.6 | Improved |
| use more rotated views (single view to 4 views) (ablation) | MVFS · NYU test | SSC mIoU (%): 51.3 -> 52.6 | Improved |
| use more rotated views (ablation) | MVFS · NYU test | SC IoU (%): 72.9 -> 73.7 | No clear effect |
| use view tokens as key/value for cross-view fusion instead of concatenated feature maps (ablation) | CVTr · NYU test | SSC mIoU (%): 51.6 -> 52.6 | Improved |
| all-for-one fusion over feature maps instead of all-fusion self-attention (ablation) | CVTr · NYU test | SSC mIoU (%): 52.3 -> 51.6 | Hurt |
| increase view-token resolution from M=75 to 144 or 245 (ablation) | CVTr · NYU test | SC IoU and SSC mIoU | Hurt |
| voxel-level supervision only versus instance-level supervision of SISNet | CVSformer · NYU test | SSC mIoU (%): 52.4 -> 52.6 | Improved |
Closest papers by meaning
- Context and Geometry Aware Voxel Transformer for Semantic Scene Completion (NeurIPS 2024 · similarity 0.91)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.90)
- SDFormer: Vision-based 3D Semantic Scene Completion via SAM-assisted Dual-channel Voxel Transformer (ICCV 2025 · similarity 0.90)
- Memory-Augmented Re-Completion for 3D Semantic Scene Completion (AAAI 2025 · similarity 0.88)
- Voxel Proposal Network via Multi-Frame Knowledge Distillation for Semantic Scene Completion (NeurIPS 2024 · similarity 0.88)
- Unleashing Network Potentials for Semantic Scene Completion (CVPR 2024 · similarity 0.87)
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.
