Reproduce “From Big to Small: Multi-Scale Local Planar Guidance for Monocular Depth Estimation” 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
Jin Han Lee, Myung-Kyu Han, Dong Wook Ko, Il Hong Suh
Hanyang University researchers developed Multi-Scale Local Planar Guidance (LPG) layers for deep convolutional networks, explicitly guiding feature map resolution recovery to improve monocular depth estimation. Their method achieved state-of-the-art performance on NYU Depth V2 and KITTI benchmarks, demonstrating enhanced accuracy across metrics, including approximately a 10% RMSE reduction on KITTI's 0-50m range.
Key insights
- Explicitly guiding feature map resolution recovery through local planar assumptions significantly enhances monocular depth estimation accuracy and detail preservation.
- Parameterizing local depth patches with just four plane coefficients offers an efficient and effective mechanism for reconstructing fine-grained depth information, outperforming generic deconvolutional upsampling.
- A multi-scale approach, where different scales learn global scene shapes (coarser) and local details or error compensation (finer), leads to improved overall depth map quality and object boundary precision.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.