New Lumera benchmark enables editable 3D scene creation from single images

Researchers have introduced Lumera, a new benchmark and pipeline for creating editable 3D scenes from single images. Lumera-2K, a dataset derived from over 2,500 Unreal Engine 5 projects, includes millions of object instances and thousands of parametric lights. The Lumera-Box and Lumera-Light components adapt vision-language models to parse object bounding boxes and light properties, enabling more detailed scene reconstruction. While Lumera-Box shows strong performance in object detection and scene layout, Lumera-Light demonstrates high recall for lights but faces challenges in precise localization and intensity estimation, highlighting areas for future research. AI

IMPACT This research could advance the creation of interactive and editable 3D environments, potentially impacting game development and virtual reality.

RANK_REASON The cluster describes a new academic paper introducing a benchmark and pipeline for 3D scene reconstruction. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• DetAny3D
• Henghaofan Zhang
• Lumera
• Lumera-2K
• Lumera-Box
• Lumera-Light
• N3D-VLM
• SpatialLM
• Unreal Engine 5
• vision-language model
• WildDet3D

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →