Reproduce “The Devil is in the Edges: Monocular Depth Estimation with Edge-aware Consistency Fusion” 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
Pengzhi Li, Yikang Ding, Haohan Wang, Chengshuai Tang, Zhiheng Li
Researchers from Tsinghua University developed ECFNet, an architecture that enhances monocular depth estimation by explicitly leveraging edge information through a hybrid edge detection strategy and consistency fusion. The method improves depth accuracy, particularly at object boundaries, achieving an 8% improvement over DPT on the NYU-v2 dataset in terms of squared relative error for edge depth.
Key insights
- Explicitly leveraging edge information is critical for accurately estimating depth at object boundaries, where discontinuities occur.
- Fusing complementary depth predictions from the original RGB image, a hybrid edge map, and an edge-highlighted image leads to more precise depth maps.
- A two-module approach, combining a Layered Fusion Module (LFM) for edge accuracy and a Depth Consistency Module (DCM) for overall structural coherence, refines the final depth output.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.