MINER: Modeling Inter- and Intra-Person Variance for Reidentification | WACV 2027

MINER is a model-agnostic framework for Clothes-Changing Person Re-Identification (CC-ReID). It combines stochastic hard mining, multi-positive contrastive learning, and an adaptive variance controller to balance inter-person discrimination and intra-person clothing invariance. MINER achieves leading Rank@1 or mAP results on **4 of 6 CC-ReID benchmarks** and podium finishes in all three tracks of the **ICPR 2026 TVRID challenge**. In this video, we explain MINER's motivation, methodology, mathematical formulation, and experimental results. Authors: Oron Nir, Eliyahu Strugo, Ariel Shamir Affiliations: Reichman University & Microsoft 🌐 Paper, code & model checkpoints: oronnir.github.io/MINER/ #WACV2027 #ComputerVision #PersonReID #DeepLearning #MINER