Reproduce “Gaussian YOLOv3: An Accurate and Fast Object Detector Using Localization Uncertainty for 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

Jiwoong Choi, Dayoung Chun, Hyun Kim, Hyuk-Jae Lee

Gaussian YOLOv3 represents each predicted bounding-box coordinate as a Gaussian distribution and uses the estimated localization uncertainty to adjust detection scores. On driving-scene datasets, it improves mAP over YOLOv3 while maintaining nearly the same inference speed; for example, KITTI mAP rises by 3.09 percentage points at 512 × 512, with speed changing from 43.57 to 43.13 FPS.

Key insights

  • Localization reliability provides information separate from whether an object is present or which class it belongs to; boxes with lower predicted uncertainty generally have higher IoU with ground truth.
  • The changes are confined to bounding-box outputs and scoring, adding an estimated 0.04% to the reported FLOPs for the specified configuration.

Summary by alphaXiv.

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