Reproduce “OpenLongTail: Generative Scaling of Long-Tail Driving Data” 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

Lulin Liu, Nuo Chen, Yan Wang, Bangya Liu, Wenyan Cong, Hezhen Hu, Boris Ivanovic, Hao Wang and 10 more

OpenLongTail, a generative data engine, converts heterogeneous, often monocular, long-tail driving videos into synchronized multi-view assets suitable for training autonomous driving policies. This framework demonstrably improves closed-loop policy robustness in rare scenarios by effectively utilizing previously incompatible real-world data.

Key insights

  • Combining refined metric-scale ego-trajectory recovery with geometry-aware synthesis is crucial for generating consistent multi-view data from fragmented monocular inputs.
  • Extrapolative view synthesis can maintain high visual fidelity and geometric consistency across views, even for largely unobserved regions, by leveraging geometric priors and a cross-view memory mechanism.
  • Generative scaling with external long-tail data sources significantly enhances autonomous driving policy robustness, but its effectiveness is influenced by the distributional alignment between source and target long-tail slices.

Summary by alphaXiv.

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