RIC: Rotate-Inpaint-Complete for Generalizable Scene Reconstruction
This work addresses the challenge of 3D scene reconstruction for applications like AR/VR and robotics, where only a single view is available, but it is incremental as it builds on existing inpainting and lifting techniques.
The paper tackles the problem of general scene reconstruction from a single view by breaking it into two steps: using a large visual language model to inpaint missing areas in rendered novel views, and then lifting these images to 3D by predicting normals instead of depth. The method outperforms multiple baselines and generalizes to novel objects and scenes, as shown through rigorous quantitative evaluation.
General scene reconstruction refers to the task of estimating the full 3D geometry and texture of a scene containing previously unseen objects. In many practical applications such as AR/VR, autonomous navigation, and robotics, only a single view of the scene may be available, making the scene reconstruction task challenging. In this paper, we present a method for scene reconstruction by structurally breaking the problem into two steps: rendering novel views via inpainting and 2D to 3D scene lifting. Specifically, we leverage the generalization capability of large visual language models (Dalle-2) to inpaint the missing areas of scene color images rendered from different views. Next, we lift these inpainted images to 3D by predicting normals of the inpainted image and solving for the missing depth values. By predicting for normals instead of depth directly, our method allows for robustness to changes in depth distributions and scale. With rigorous quantitative evaluation, we show that our method outperforms multiple baselines while providing generalization to novel objects and scenes.