Reproduce “Unsupervised Monocular Depth Estimation with Left-Right Consistency” 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

Clément Godard, Oisin Mac Aodha, Gabriel J. Brostow

Researchers at UCL developed an unsupervised learning method for monocular depth estimation that leverages binocular stereo images and a novel left-right disparity consistency loss. The approach achieved state-of-the-art performance on the KITTI benchmark, surpassing supervised methods trained with ground truth depth.

Key insights

  • High-quality monocular depth estimation is achievable without explicit ground truth depth data by exploiting self-supervision from binocular stereo image pairs.
  • The left-right disparity consistency loss is crucial for enforcing geometric coherence, leading to robust and accurate disparity maps, particularly at object boundaries.
  • Unsupervised learning, when effectively designed, can outperform supervised methods on challenging benchmarks, validating its potential as a powerful paradigm.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.