Research map / 3D vision / 3D generation
Pano3DComposer: Feed-Forward Compositional 3D Scene Generation from Single Panoramic Image
Zidian Qiu, Ancong Wu
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 pseudo-geometry distillation loss (ablation) | Alignment-VGGT · 3D-FRONT test set, alignment loss ablation | CD-S: 0.8688 -> 0.1266 | Improved |
| add mask regularization loss (ablation) | Alignment-VGGT · 3D-FRONT test set, alignment loss ablation | CD-S: 0.1266 -> 0.1120 | Improved |
| add mask regularization loss (ablation) | Alignment-VGGT · 3D-FRONT test set, alignment loss ablation | IoU-B: 0.4670 -> 0.4818 | Improved |
| add camera intrinsics and extrinsics as input (ablation) | Alignment-VGGT · 3D-FRONT test set, alignment loss ablation | CD-S: 0.1850 -> 0.1120 | Improved |
| freeze DINO backbone during fine-tuning (ablation) | Alignment-VGGT · 3D-FRONT test set, VGGT fine-tuning ablation | CD-S: 0.1883 -> 0.1236 | Improved |
| also freeze frame attention layers (ablation) | Alignment-VGGT · 3D-FRONT test set, VGGT fine-tuning ablation | CD-S: 0.1236 -> 0.0787 | Improved |
| also freeze global attention layers (ablation) | Alignment-VGGT · 3D-FRONT test set, VGGT fine-tuning ablation | CD-S: 0.0787 -> 0.1120 | Hurt |
| add coarse-to-fine iterative refinement | Pano3DComposer · 3D-FRONT test set | IoU-B: 0.5679 -> 0.5699 | Improved |
| replace differentiable optimization with feed-forward predictor | alignment · 3D-FRONT test set | CD-S: 0.1059 -> 0.0787 | Improved |
| replace ICP with feed-forward predictor | alignment · 3D-FRONT test set | CD-S: 0.2483 -> 0.0787 | Improved |
| replace SceneGen with Pano3DComposer | scene composition · 3D-FRONT test set | IoU-B: 0.1124 -> 0.5679 | Improved |
| replace SceneGen with Pano3DComposer | scene composition · 3D-FRONT test set | inference time: 63s -> 20s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SceneGen — SceneGen: Learning To Generate Realistic Traffic Scenes (CVPR 2021) · better on 4 of 4 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
- PanoWorld: Geometry-Consistent Panoramic Video World Modeling (NeurIPS 2026 · similarity 0.89)
- Scene4U: Hierarchical Layered 3D Scene Reconstruction from Single Panoramic Image for Your Immerse Exploration (CVPR 2025 · similarity 0.89)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.88)
- Splatter-360: Generalizable 360 Gaussian Splatting for Wide-baseline Panoramic Images (CVPR 2025 · similarity 0.88)
- DiffPano: Scalable and Consistent Text to Panorama Generation with Spherical Epipolar-Aware Diffusion (NeurIPS 2024 · similarity 0.87)
- Structured 3D Latents for Scalable and Versatile 3D Generation (CVPR 2025 · 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.
