UPAR Challenge 2027:
Attributes, Retrieval, and Pose Estimation
RWS introduces a new edition of the UPAR Challenge, extending the previous editions on Pedestrian Attribute Recognition (PAR) and Attribute-Based Person Retrieval (ABPR) towards unified pedestrian understanding. The challenge builds upon the UPAR benchmark by incorporating Human Pose Estimation (HPE), enabling evaluation of pedestrian appearance and body structure understanding across diverse surveillance domains. The challenge consists of three tracks.
The first track continues PAR under challenging domain shifts, while the second extends ABPR to cross-dataset pedestrian retrieval.
The third track introduces surveillance-oriented HPE, evaluating keypoint localization under variations in viewpoint, resolution, illumination, and occlusion using the UPAR-Pose benchmark. A development phase for method validation and a test phase with submission limits to avoid overfitting
are included. Baseline methods for all tracks will be provided.
Challenge Schedule:
This challenge is organized by Mickael Cormier and Andreas Specker. The challenge will be hosted on CodaBench, starting September 15th.
