Mira-Scene: Pixel-Aligned Layouts for Generative 3D Scene
Mira-Scene is a new 3D scene reconstruction framework that improves object layout accuracy using dense canonical coordinate maps and multimodal diffusion techniques.
This paper presents Mira-Scene, a 3D scene reconstruction framework that improves object layout accuracy using dense canonical coordinate maps and multimodal diffusion techniques.
- Mira-Scene is a 3D scene reconstruction framework for object layout generation
- It uses dense canonical coordinate maps to improve layout accuracy over SAM3D
- Experiments show 39.8% and 16.5% improvement over baselines
- Products
- Mira-Scene
- Countries
- IndoorOutdoorSyntheticIn-the-wild
- AI models
- SAM3D
From the original article
1 min read at huggingface.co
Single-image 3D object generation can now produce high-fidelity assets, yet accurately placing them into a coherent scene layout remains an open challenge. A central difficulty lies in how object layout is represented. Holistic methods absorb placement into a scene-level generation process, sacrificing object-level detail.
By Yang-Tian Sun, Tianjia Liu, Zehuan Huang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.