Research map / 3D vision / 3D generation
Omni-View: Unlocking How Generation Facilitates Understanding in Unified 3D Model based on Multiview images
JiaKui Hu, Shanshan Zhao, Qing-Guo Chen, Xuerui Qiu, Jialun Liu, Zhao Xu, Weihua Luo, Kaifu Zhang, Yanye Lu
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 generation training (texture module) to understanding model (ablation) | Omni-View (BAGEL-7B-FT) · VSI-Bench Appr. Order | accuracy: 43.1 -> 47.2 | Improved |
| add geometry module (depth) on top of texture (ablation) | Omni-View · VSI-Bench Rel. Dist. | accuracy: 53.2 -> 63.0 | Improved |
| add geometry module (depth) (ablation) | Omni-View · VSI-Bench Abs. Dist. | accuracy: 44.6 -> 44.9 | No clear effect |
| separate texture and geometry modules instead of unified architecture (ablation) | Omni-View · VSI-Bench Rel. Dist. | accuracy: 63.8 -> 65.9 | Improved |
| use autoregressive instead of bidirectional generation (ablation) | Omni-View · VSI-Bench Abs. Dist. | accuracy: 40.6 -> 46.4 | Improved |
| use autoregressive instead of bidirectional generation (ablation) | Omni-View · SQA3D | EM: 57.2 -> 59.2 | Improved |
| use dense-to-sparse reference schedule vs dense only (ablation) | Omni-View · ScanQA CIDEr | CIDEr: 94.7 -> 103.0 | Improved |
| use dense-to-sparse vs Ross3D random mask (ablation) | Omni-View · ScanQA CIDEr | CIDEr: 98.9 -> 103.0 | Improved |
| add dense-only references (ablation) | Omni-View · VSI-Bench Obj. Size | accuracy: dense 60.1 vs sparse 65.5 | Mixed |
| add stage 2 RGBDP fine-tuning (ablation) | Omni-View · Re10k NVS single view | LPIPS: 0.265 -> 0.114 | Improved |
| compare to Voyager-13B on NVS pixel fidelity | Omni-View · Re10k NVS single view | PSNR: 23.12 -> 23.22 | No clear effect |
| compare to 3D-input methods on grounding | Omni-View · ScanRefer | Acc@0.25: 62.6 -> 50.8 | Hurt |
Closest papers by meaning
- Generation Enhances Understanding in Unified Multimodal Models via Multi-Representation Generation (ICML 2026 · similarity 0.86)
- Inst3D-LMM: Instance-Aware 3D Scene Understanding with Multi-modal Instruction Tuning (CVPR 2025 · similarity 0.86)
- UniUGG: Unified 3D Understanding and Generation via Geometric-Semantic Encoding (ICLR 2026 · similarity 0.86)
- Cross-Modal 3D Representation with Multi-View Images and Point Clouds (CVPR 2025 · similarity 0.85)
- A Unified Framework for 3D Scene Understanding (NeurIPS 2024 · similarity 0.85)
- GenSpace: Benchmarking Spatially-Aware Image Generation (NeurIPS 2025 · similarity 0.85)
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.
