Research map / 3D vision / 3D generation
Hierarchical Temporal Context Learning for Camera-based Semantic Scene Completion
Bohan Li, Jiajun Deng, Wenyao Zhang, Zhujin Liang, Dalong Du, Xin Jin, Wenjun Zeng
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 |
|---|---|---|---|
| use temporal feature volume instead of cost volume (ablation) | HTCL-S · SemanticKITTI val | mIoU: 16.02 -> 17.13 | Improved |
| add scale-aware isolation to cosine affinity (ablation) | HTCL-S CPA · SemanticKITTI val | mIoU: 15.18 -> 17.13 | Improved |
| add multi-group context generation (ablation) | HTCL-S CPA · SemanticKITTI val | mIoU: 15.26 -> 17.13 | Improved |
| add affinity weights to dynamic refinement (ablation) | HTCL-S ADR · SemanticKITTI val | mIoU: 14.65 -> 17.13 | Improved |
| use deformable convolutions instead of normal convolutions (ablation) | HTCL-S ADR · SemanticKITTI val | mIoU: 15.14 -> 17.13 | Improved |
| add learnable coefficient in weighted voxel attention (ablation) | HTCL-S WVA · SemanticKITTI val | mIoU: 15.85 -> 17.13 | Improved |
| use voxel cross-attention instead of concatenation (ablation) | HTCL-S WVA · SemanticKITTI val | mIoU: 15.98 -> 17.13 | Improved |
| use more than 3 previous temporal frames (ablation) | HTCL-S · SemanticKITTI val | mIoU | No clear effect |
| use 5 context groups instead of 3 (ablation) | HTCL-S CPA · SemanticKITTI val | mIoU: 17.13 -> 17.21 | No clear effect |
| use 5 deformable levels instead of 3 (ablation) | HTCL-S ADR · SemanticKITTI val | mIoU: 17.13 -> 17.18 | No clear effect |
| use 3 deformable levels instead of 1 (ablation) | HTCL-S ADR · SemanticKITTI val | mIoU: 16.51 -> 17.13 | Improved |
| give stacked-temporal-stereo variants of single-frame baselines the same temporal inputs | MonoScene, TPVFormer, OccFormer · SemanticKITTI val | mIoU: 12.96, 13.21, 13.57 vs HTCL-S 17.13 | No clear effect |
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.
- OccFormer — OccFormer: Dual-path Transformer for Vision-based 3D Semantic Occupancy Prediction (ICCV 2023) · better on 1 of 1 shared result
- MonoScene — MonoScene: Monocular 3D Semantic Scene Completion (CVPR 2022) · better on 1 of 1 shared result
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 Temporal 3D Semantic Scene Completion via Optical Flow Guidance (NeurIPS 2025 · similarity 0.90)
- Context and Geometry Aware Voxel Transformer for Semantic Scene Completion (NeurIPS 2024 · similarity 0.89)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.89)
- VLScene: Vision-Language Guidance Distillation for Camera-Based 3D Semantic Scene Completion (AAAI 2025 · similarity 0.88)
- Semi-supervised 3D Semantic Scene Completion with 2D Vision Foundation Model Guidance (AAAI 2025 · similarity 0.88)
- IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals (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.
