Research map / Image generation / Fast diffusion sampling
Merging and Splitting Diffusion Paths for Semantically Coherent Panoramas
Fabio Quattrini, Vittorio Pippi, Silvia Cascianelli, Rita Cucchiara
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 |
|---|---|---|---|
| apply MAD in all U-Net blocks instead of never (ablation) | MAD on LDM · 512x3072 panoramas, LDM | I-LPIPS: 0.64 -> 0.53 | Improved |
| apply MAD only in the mid (bottleneck) block (ablation) | MAD on LDM · 512x3072 panoramas, LDM | FID: 34.51 -> 34.80 | No clear effect |
| apply MAD in all blocks instead of never (ablation) | MAD on LDM · 512x3072 panoramas, LDM | FID: 34.51 -> 61.76 | Hurt |
| increase number of MAD application steps tau from 0 to 15 (ablation) | MAD on LDM · 512x3072 panoramas, LDM | I-LPIPS: 0.64 -> 0.56 | Improved |
| increase tau from 15 to 50 (apply MAD at every step) (ablation) | MAD on LDM · 512x3072 panoramas, LDM | FID: 48.52 -> 98.01 | Hurt |
| combine MAD with attention scaling (Attn-S) (ablation) | MAD on LDM · 512x3072 panoramas, LDM, standard prompts | mCLIP: 32.03 -> 32.17 | No clear effect |
| use MAD instead of direct inference at 512x3072 | Stable Diffusion LDM · 512x3072 panoramas, standard prompts | FID: 87.64 -> 48.52 | Improved |
| use MAD instead of direct inference with 1 LCM step | LCM · 512x3072 panoramas, LCM 1 step | FID: 67.54 -> 72.43 | Hurt |
| use MAD instead of direct inference with 2 LCM steps | LCM · 512x3072 panoramas, LCM 2 steps | FID: 49.98 -> 35.69 | Improved |
| use MAD instead of MultiDiffusion and SyncDiffusion | panorama generation · user study, 6 prompts | human preference | Improved |
Closest papers by meaning
- CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation (ICLR 2025 · similarity 0.84)
- Taming Stable Diffusion for Text to 360 Panorama Image Generation (CVPR 2024 · similarity 0.83)
- Structure Matters: Tackling the Semantic Discrepancy in Diffusion Models for Image Inpainting (CVPR 2024 · similarity 0.81)
- HORIZON: High-Resolution Semantically Controlled Panorama Synthesis (AAAI 2024 · similarity 0.81)
- DiffPano: Scalable and Consistent Text to Panorama Generation with Spherical Epipolar-Aware Diffusion (NeurIPS 2024 · similarity 0.81)
- SemCity: Semantic Scene Generation with Triplane Diffusion (CVPR 2024 · similarity 0.81)
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.
