Reproduce “CODA: A Real-World Road Corner Case Dataset for Object Detection in Autonomous Driving” 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

Kaican Li, Kai Chen, Haoyu Wang, Lanqing Hong, Chaoqiang Ye, Jianhua Han, Yukuai Chen, Wei Zhang and 5 more

CODA introduces a real-world dataset comprising over 6000 object-level corner cases across 30+ categories for autonomous driving object detection. It quantifies that state-of-the-art detectors exhibit a 30-50% drop in Average Recall on these challenging scenarios and presents COPG, an efficient pipeline for discovering such rare cases.

Key insights

  • State-of-the-art object detectors exhibit a significant performance gap, with an average 30-50% drop in Average Recall, when evaluated on real-world road corner cases compared to their performance on standard datasets.
  • Even with broader exposure to diverse object categories during pre-training, overall performance on corner cases remains very low, indicating that current models struggle with the inherent variability of real-world driving.
  • The COPG pipeline effectively identifies challenging corner cases and significantly reduces human annotation effort, demonstrating its potential for efficient construction of larger and more diverse datasets for robust perception.

Summary by alphaXiv.

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