Reproduce “ACT-R: Adaptive Camera Trajectories for Single View 3D Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yizhi Wang, Mingrui Zhao, Hao Zhang
ACT-R, developed at Simon Fraser University, introduces an adaptive camera trajectory planning method for single-view 3D reconstruction, strategically generating novel views to reveal occluded object structures. This approach significantly improves geometric fidelity and occlusion handling, outperforming state-of-the-art methods on the Google Scanned Objects dataset by enhancing both 3D and 2D reconstruction metrics.
Key insights
- Adaptive planning of camera trajectories, specifically designed to maximize visibility of occluded regions, leads to substantial improvements in single-view 3D reconstruction quality and completeness.
- Combining insights from multi-slice methods (for identifying occlusions) with multi-view generation (for visible geometry) provides a robust understanding of object geometry from a single image.
- Generating a sequence of views along a planned trajectory, rather than an unordered set, leverages temporal consistency from video generation models, enhancing 3D coherence and reducing artifacts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.