AI Paper Atlas

Research map / Image generation / Editing and style transfer

World-Shaper: A Unified Framework for 360° Panoramic Editing

Dong Liang, Yuhao LIU, Jinyuan Jia, Youjun Zhao, Rynson W. H. Lau

ICML 2026 · Image generation › Editing and style transfer · approach: Localized and attribute-specific editing · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add distortion-aware attention modulation and layered shape loss together (ablation)World-Shaper · PEBench panorama editingVIEScore Overall: 5.32 -> 8.01Improved
add distortion-aware attention modulation and layered shape loss together (ablation)World-Shaper · PEBench panorama editingFID: 58.32 -> 45.25Improved
add distortion-aware attention modulation (DAAM) alone (ablation)World-Shaper · PEBench panorama editingVIEScore Overall: 5.32 -> 7.45Improved
add layered shape loss (LSL) alone (ablation)World-Shaper · PEBench panorama editingVIEScore Overall: 5.32 -> 7.38Improved
add LSL on top of DAAM (ablation)World-Shaper · PEBench panorama editingFID: 53.82 -> 45.25Improved
add Stage 1 global panorama generation pretraining before Stage 3 editing (ablation)World-Shaper curriculum · PEBench panorama editingFID: 52.33 -> 45.25 (with Stage 2 present)Improved
add Stage 2 localized controllable generation before Stage 3 editing (ablation)World-Shaper curriculum · PEBench panorama editingCLIP direction score x100: 18.36 -> 19.62 (with Stage 1 present)Improved
add both Stage 1 and Stage 2 versus training Stage 3 directly from Qwen-Image-Edit weights (ablation)World-Shaper curriculum · PEBench panorama editingVIEScore Overall: 7.18 -> 8.01Improved
use ERP-native World-Shaper instead of perspective-based or cube-map editing baselinespanorama editing · PEBench panorama editingVIEScore Overall: 5.66 (best baseline Qwen) -> 8.01Improved
use World-Shaper instead of the best baseline in user rankingpanorama editing · PEBench user study (72 participants)average rank, image quality: 2.37 -> 1.19Improved
enhance editing prompts with GPT-5 for all methodsall editing methods incl. World-Shaper · PEBench automatic metricsVIEScore Overall: +0.11 for World-Shaper (8.01 -> 8.12); +0.17 for QwenImproved
keep the ranking of methods after prompt enhancementWorld-Shaper vs baselines · PEBench with enhanced promptsFID: 45.25 -> 44.65 (-0.60)No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy