About the Challenge

The UIoU Dark Zurich challenge focuses on uncertainty-aware semantic nighttime image segmentation. Participants are tasked with developing models that not only segment nighttime scenes accurately but also quantify the uncertainty of their predictions — a critical capability for safety-sensitive applications such as autonomous driving and surveillance.

The Dark Zurich Dataset

The challenge uses the Dark Zurich dataset, first presented in the ICCV 2019 paper "Guided Curriculum Model Adaptation and Uncertainty-Aware Evaluation for Semantic Nighttime Image Segmentation." The dataset contains a total of 8779 images captured across three conditions:

• Nighttime — scenes under low and artificial illumination
• Twilight — transitional lighting between day and night
• Daytime — corresponding daytime images used as reference for reliable label generation

Evaluation Framework

A novel evaluation framework leverages daytime images to produce reliable ground-truth labels for semantically corresponding nighttime scenes. This enables a principled assessment of both segmentation accuracy and prediction uncertainty under adverse lighting.