Reproduce “DIODE: A Dense Indoor and Outdoor DEpth Dataset” 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
Igor Vasiljevic, Nick Kolkin, Shanyi Zhang, Ruotian Luo, Haochen Wang, Falcon Z. Dai, Andrea F. Daniele, Mohammadreza Mostajabi and 3 more
DIODE is a dense indoor and outdoor depth dataset, comprising over 25,000 high-resolution (768x1024) RGBD images from both indoor and outdoor environments, captured using a consistent, high-accuracy FARO Focus S350 laser scanner. This dataset uniquely offers empirical depth return densities of 99.6% indoors and 66.9% outdoors with ±1mm precision, enabling the training of single models that generalize more effectively across diverse scenes for monocular depth estimation.
Key insights
- A unified, high-fidelity sensor suite is crucial for collecting diverse indoor and outdoor depth data that enables the training of generalizable depth estimation models.
- Dense and accurate depth information across varying ranges and environments can be captured and processed to support sophisticated 2.5D and 3D vision tasks.
- Training a single monocular depth estimation model on a combined indoor and outdoor dataset (DIODE/All) leads to better overall generalization across both domains compared to training on domain-specific subsets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.