Building Rome from a Single Image
A redesigned object generator builds complete indoor and outdoor 3D meshes from one image, beating prior baselines.
The paper redesigns an object-centric 3D generator, using Trellis 2 as an example, to reconstruct complete indoor and outdoor scene meshes from a single image, including unobserved surfaces. Adaptive chunks scale with camera distance, and lifted image features distinguish free space, observed surfaces, and unobserved regions. The authors synthesize about 4,000 outdoor scenes to supplement mostly indoor datasets. On Tanks and Temples, ScanNet++, and in-the-wild images, the method beat baselines in geometric accuracy and perceptual quality.
- Adapts an object-centric generator such as Trellis 2 to full scenes.
- Distance-adaptive chunks preserve nearby detail and cover distant structures.
- About 4,000 synthetic outdoor scenes expand largely indoor training data.
- Outperforms baselines on Tanks and Temples, ScanNet++, and in-the-wild images.
- Products
- Trellis 2
From the original article
1 min read at huggingface.co
Single-image scene generation aims to produce a complete 3D scene mesh from a single image, including surfaces the camera did not observe. While pretrained 3D object generators encode a strong shape prior, they are mainly designed for isolated objects in a fixed canonical volume and focus mostly on indoor scenes, since diverse 3D data for outdoor scenes are quite limited.
By Jiraphon Yenphraphai, Fang Li, Tianshuo Xu · 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.