Reproduce “EvCenterNet: Uncertainty Estimation for Object Detection using Evidential Learning” 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

Monish R. Nallapareddy, Kshitij Sirohi, Paulo L. J. Drews-Jr, Wolfram Burgard, Chih-Hong Cheng, Abhinav Valada

EvCenterNet introduces a 2D object detection framework that directly provides both classification and regression uncertainties in a single forward pass, leveraging evidential deep learning. This approach offers competitive detection precision, high computational efficiency, and robust generalization to out-of-distribution data, critical for safety-critical applications like autonomous driving.

Key insights

  • Evidential learning can be effectively adapted for object detection to provide direct, sampling-free classification and regression uncertainties.
  • Specific loss function modifications, such as class-balanced weighting and one-sided focal loss, are essential for evidential learning with sparse object detection outputs like heatmaps and for maintaining detection accuracy.
  • Actively leveraging predicted uncertainties during training through regularization losses significantly enhances the overall detection performance and model robustness.

Summary by alphaXiv.

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