Reproduce “RIDE: Relocalization-Informed Depth Estimation with 3D Gaussian Splatting” 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
Jiarong Lian, Zhe Xiao, Zhaoyang Zhang, Wei Li, Ruizhi Chen
RIDE turns relocalization matches into sparse metric depth anchors for calibrating a dense video-depth model. Combining robust scale estimation, spatial correction, and flow-guided temporal memory, it improves temporal consistency and real-robot depth accuracy without deployment-time depth sensing or test-time fine-tuning.
Key insights
- Relocalization correspondences provide supervision for dense metric depth as well as pose, creating a bridge between localization and depth perception without adding a depth sensor.
- Local correction is necessary beyond global scale calibration: the global-only ablation has 0.2173 absolute relative error, versus 0.1178 after adding spatial correction.
- Temporal memory and flow-based alignment are important for stability; shuffling historical states raises temporal geometric error from 0.0354 to 0.0754.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.