Reproduce “Incremental Object Database: Building 3D Models from Multiple Partial Observations” 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
Fadri Furrer, Tonci Novkovic, Marius Fehr, Abel Gawel, Margarita Grinvald, Torsten Sattler, Roland Siegwart, Juan Nieto
Researchers at ETH Zurich developed an incremental system that automatically builds a database of 3D object models from multiple partial observations using RGB-D data. The system integrated 48 non-planar raw segments into 27 distinct object models, demonstrating the completion of objects such as a cow from three raw segments and a round chair from ten, leading to more complete scene reconstructions.
Key insights
- Maintaining a complete history of segment labels and their counts within the GSM's label volume enhances the robustness of label fusion against noise and allows for lossless merging of segments.
- A multi-stage registration and verification process, combining descriptor matching, coarse RANSAC, and fine ICP, enables the unsupervised discovery and alignment of previously uncataloged objects.
- Merging Truncated Signed Distance Field (TSDF) representations from multiple partial observations of the same object significantly improves the completeness and accuracy of 3D object models and overall scene reconstructions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.