Research map / 3D vision / Gaussian splatting
TGSFormer: Scalable Temporal Gaussian Splatting for Embodied Semantic Scene Completion
Rui Qian, Haozhi Cao, Tianchen Deng, Tianxin Hu, Weixiang Guo, Shenghai Yuan, Lihua Xie
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 Confidence-aware Voxel Fusion with confidence weighting (ablation) | TGSFormer · CAVF ablation (embodied) | mIoU: 41.84 -> 50.55 | Improved |
| add confidence weighting to voxel fusion (0.12m) (ablation) | TGSFormer CAVF · CAVF ablation (embodied) | mIoU: 45.76 -> 50.55 | Improved |
| add CAVF voxel fusion (ablation) | TGSFormer · CAVF ablation (embodied) | Gaussian number per frame: 4086.0 -> 859.0 | Improved |
| use voxel size 0.14m instead of 0.12m (ablation) | TGSFormer CAVF · CAVF ablation (embodied) | mIoU: 50.55 -> 49.88 | Hurt |
| add dual cross attention with confidence modulation (ablation) | TGSFormer DTE · Temporal encoder ablation (embodied) | mIoU: 49.35 -> 49.70 | Improved |
| use single cross attention instead of none (ablation) | TGSFormer DTE · Temporal encoder ablation (embodied) | mIoU: 49.35 -> 48.52 | Hurt |
| add dual cross attention without confidence (ablation) | TGSFormer DTE · Temporal encoder ablation (embodied) | mIoU: 49.35 -> 48.29 | Hurt |
| supervise intermediate stages 1 and 2 instead of stage 2 only (ablation) | TGSFormer · Supervision ablation (embodied) | mIoU: 48.04 -> 49.70 | Improved |
| supervise all stages 0,1,2 (ablation) | TGSFormer · Supervision ablation (embodied) | mIoU: 49.70 -> 48.63 | Hurt |
| modulate attention with confidence on value and output (ablation) | TGSFormer CCA · CCA modulation ablation (monocular) | mIoU: 54.03 -> 54.41 | Improved |
| modulate confidence on query only (ablation) | TGSFormer CCA · CCA modulation ablation (monocular) | mIoU: 54.03 -> 53.55 | Hurt |
| fine-tune Depth Anything V2 for Gaussian initialization (ablation) | TGSFormer · Gaussian init ablation | mIoU: 39.50 -> 54.41 | Improved |
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.
- SplatSSC — SplatSSC: Decoupled Depth-Guided Gaussian Splatting for Semantic Scene Completion (AAAI 2026) · better on 4 of 4 shared results
- EmbodiedOcc++ — EmbodiedOcc: Embodied 3D Occupancy Prediction for Vision-based Online Scene Understanding (ICCV 2025) · better on 3 of 3 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
- Learning Efficient Fuse-and-Refine for Feed-Forward 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.88)
- VoxelSplat: Dynamic Gaussian Splatting as an Effective Loss for Occupancy and Flow Prediction (CVPR 2025 · similarity 0.88)
- IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals (NeurIPS 2025 · similarity 0.88)
- Learning Temporal 3D Semantic Scene Completion via Optical Flow Guidance (NeurIPS 2025 · similarity 0.88)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.88)
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 2025 · similarity 0.88)
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.
