Research map / 3D vision / 3D generation
Lifting Unlabeled Internet-level Data for 3D Scene Understanding
Yixin Chen, Yaowei Zhang, Huangyue Yu, Junchao He, Yan Wang, Jiangyong Huang, Hongyu Shen, Junfeng Ni, Shaofei Wang, Baoxiong Jia, Song-Chun Zhu, Siyuan Huang
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 |
|---|---|---|---|
| Pretrain on SceneVerse++ then finetune on ScanNet | SpatialLM · ScanNet detection | F1@0.25: 38.0 -> 58.6 | Improved |
| Pretrain on SceneVerse++ instead of synthetic data (zero-shot) | SpatialLM · ScanNet detection | F1@0.25: 29.0 -> 30.9 | Improved |
| Pretrain on SceneVerse++ instead of synthetic data (zero-shot) | SpatialLM · ARKitScenes detection | F1@0.5: 21.2 -> 20.7 | No clear effect |
| Pretrain on SceneVerse++ only, zero-shot | Mask3D · ScanNet instance segmentation | AP25: 36.1 -> 15.4 | Hurt |
| Pretrain on SceneVerse++ then finetune on ScanNet | Mask3D · ScanNet instance segmentation | AP25: 36.1 -> 38.5 | Improved |
| Train on SceneVerse++ spatial QA | Qwen2.5-VL-3B · VSI-Bench full set | Average score: 27.9 -> 42.8 | Improved |
| Train on SceneVerse++ spatial QA | Qwen2.5-VL-7B · VSI-Bench full set | Average score: 36.6 -> 46.4 | Improved |
| Combine SceneVerse++ with SN/SN++ data (vs SN/SN++ only) | Qwen2.5-VL-3B · VSI-Bench ARKit subset | Average score: 49.0 -> 51.3 | Improved |
| Train on SceneVerse++ instead of SN/SN++ | Qwen2.5-VL-7B · VSI-Bench ARKit subset | Average score: 48.8 -> 49.1 | No clear effect |
| Pretrain on SceneVerse++ then finetune on R2R (vs R2R only) | LLaVA-Video · R2R | Success rate: 0.088 -> 0.228 | Improved |
| Pretrain then finetune instead of mixing SceneVerse++ with R2R | LLaVA-Video · R2R | Success rate: 0.188 -> 0.228 | Improved |
| Add trajectory refinement (ablation) | SceneVerse++ VLN data · R2R | Success rate: 0.177 -> 0.228 | Improved |
Closest papers by meaning
- A Unified Framework for 3D Scene Understanding (NeurIPS 2024 · similarity 0.89)
- UniUGG: Unified 3D Understanding and Generation via Geometric-Semantic Encoding (ICLR 2026 · similarity 0.89)
- GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models (ICLR 2026 · similarity 0.88)
- PartNeXt: A Next-Generation Dataset for Fine-Grained and Hierarchical 3D Part Understanding (NeurIPS 2025 · similarity 0.88)
- Lexicon3D: Probing Visual Foundation Models for Complex 3D Scene Understanding (NeurIPS 2024 · similarity 0.87)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 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.
