Research map / Video / Video generation
Generation Models Know Space: Unleashing Implicit 3D Priors for Scene Understanding
Xianjin Wu, Dingkang Liang, Tianrui Feng, Kui Xia, Yumeng Zhang, Xiaofan Li, Xiao Tan, Xiang Bai
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 Wan2.1 video-diffusion features with adaptive gated fusion | Video-3D LLM · ScanRefer | Acc@0.5: 51.7 -> 56.2 | Improved |
| add generative features with gated fusion | Video-3D LLM · SQA3D | EM: 58.6 -> 61.3 | Improved |
| add generative features with gated fusion | Video-3D LLM · Scan2Cap | CIDEr@0.5: 83.8 -> 83.2 | Hurt |
| add generative features with gated fusion | Qwen2.5VL-7B · VSI-Bench | avg accuracy: 48.9 -> 50.5 | Improved |
| add generative features with gated fusion | OpenVLA-OFT · LIBERO | avg success rate: 97.0 -> 97.3 | Improved |
| add generative features | OpenVLA-OFT · Click Bell (real robot) | success rate: 40% -> 55% | Improved |
| use generative features only, dropping semantic features (ablation) | VEGA-3D · Scan2Cap | CIDEr@0.5: 83.2 -> 25.2 | Hurt |
| adaptive gated fusion instead of Add (ablation) | VEGA-3D · ScanRefer | Acc@0.25: 61.5 -> 63.2 | Improved |
| adaptive gated fusion instead of Add (ablation) | VEGA-3D · SQA3D | EM: 61.8 vs 61.3 | No clear effect |
| token-level gate instead of channel-level gate (ablation) | VEGA-3D · ScanRefer | Acc@0.25: 61.8 -> 63.2 | Improved |
| cross-attention fusion instead of gated fusion (ablation) | VEGA-3D · Scan2Cap | CIDEr@0.5: 83.2 -> 48.8 | Hurt |
| DiT-based Wan2.1 encoder instead of UNet-based Stable Video Diffusion (ablation) | VEGA-3D · ScanRefer | Acc@0.25: 61.3 -> 63.2 | Improved |
Closest papers by meaning
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · similarity 0.90)
- Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering (NeurIPS 2026 · similarity 0.89)
- Towards Physical Understanding in Video Generation: A 3D Point Regularization Approach (NeurIPS 2025 · similarity 0.89)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.88)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.88)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · 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.
