Research map / Image generation / Text-to-image diffusion
What Makes for Text to 360-degree Panorama Generation with Stable Diffusion?
Jinhong Ni, Chang-Bin Zhang, Qiang Zhang, Jing Zhang
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 |
|---|---|---|---|
| train only W_v instead of only W_q (ablation) | LoRA fine-tuned Stable Diffusion 2 · Matterport3D panorama (512x1024) | FAED: 10.86 -> 8.66 | Improved |
| train only W_o instead of only W_k (ablation) | LoRA fine-tuned Stable Diffusion 2 · Matterport3D panorama (512x1024) | FID: 67.63 -> 52.17 | Improved |
| freeze W_q and W_k (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FID: 50.40 -> 48.62 | Improved |
| freeze W_q and W_k (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FAED: 7.90 -> 7.99 | No clear effect |
| replace LoRA with MoE on W_o (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FAED: 7.90 -> 7.21 | Improved |
| replace LoRA with MoE on W_o (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FID: 50.40 -> 48.83 | Improved |
| freeze W_q,W_k and use MoE on W_o together (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FAED: 7.90 -> 5.90 | Improved |
| add deformable attention before W_o LoRA (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FAED: 7.90 -> 9.78 | Hurt |
| double LoRA rank on W_o from 4 to 8 (ablation) | Pano Only baseline · Matterport3D panorama (512x1024) | FAED: 7.90 -> 8.34 | Mixed |
| scale MoE experts n from 4 to 8 (k=2) (ablation) | UniPano · Matterport3D panorama (512x1024) | FID: 46.47 -> 47.10 | Hurt |
| increase top-k experts from 2 to 4 (n=8) (ablation) | UniPano · Matterport3D panorama (512x1024) | FAED: 6.17 -> 7.31 | Hurt |
| use uni-branch UniPano instead of dual-branch PanFusion | PanFusion · Matterport3D panorama (512x1024) | peak GPU memory (GB): 60.12 -> 32.59 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MVDiffusion — MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion (NeurIPS 2023) · better on 1 of 2 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
- Taming Stable Diffusion for Text to 360 Panorama Image Generation (CVPR 2024 · similarity 0.93)
- DiffPano: Scalable and Consistent Text to Panorama Generation with Spherical Epipolar-Aware Diffusion (NeurIPS 2024 · similarity 0.92)
- CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation (ICLR 2025 · similarity 0.91)
- Conditional Panoramic Image Generation via Masked Autoregressive Modeling (NeurIPS 2025 · similarity 0.89)
- ViewPoint: Panoramic Video Generation with Pretrained Diffusion Models (NeurIPS 2025 · similarity 0.89)
- PanoWan: Lifting Diffusion Video Generation Models to 360$^\circ$ with Latitude/Longitude-aware Mechanisms (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.
