Prototype learning improves object recognition between similar identities
Learning Discriminative Prototypes via Boundary Repulsion MixUp for Object Re-Identification
Computer Vision and Pattern Recognition
Summary
Object re-identification means teaching a computer to recognize the same person or vehicle in different images, even if they look different because of lighting or angle. The authors created a method that helps the computer better tell apart very similar-looking people or cars by learning a 'prototype' for each identity and focusing on the tricky cases where identities are often confused. They introduced a way to prevent the model from mixing up those confusing identities during training, which helps the computer keep clear boundaries between them. This approach was tested on common benchmarks and showed better results compared to previous methods.
What this means in practice
- •For security system developers: Improve accuracy in matching people across different camera views by reducing confusion between visually similar identities.
- •For automated vehicle tracking teams: Enhance vehicle re-identification in traffic monitoring by maintaining clearer distinctions between visually similar vehicles.
Authors
Athena Psalta, Vasileios Tsironis, Konstantinos Karantzalos
Abstract
Object re-identification requires embeddings that are robust to intra-class variation in viewpoint, illumination, and occlusion, yet preserve sharp separation between visually similar identities. The hardest errors arise precisely where confusable identities meet, a region that pairwise losses shape only implicitly. We present PBR-ReID, a prototype-based framework that makes this inter-class geometry an explicit learning target. Each identity is represented by a learnable prototype jointly optimized with a DiNOv3 backbone and the resulting prototype similarity matrix reveals, at every training step, which identities are most confusable. On this signal we build Boundary Repulsion MixUp (BRM), a geometry-aware regularizer that converts standard MixUp's weakness into a discriminative signal. Rather than interpolating samples across confusable identities, which blurs the boundaries that matter most, BRM constructs an explicit boundary proxy between each confusable pair and repels their features away from it, converting boundary ambiguity into a discriminative signal. We adopt a two-stage training protocol consisting of a prototype warmup phase followed by EMA-stabilized joint fine-tuning. Extensive experiments on person and vehicle ReID benchmarks demonstrate strong performance, with thorough ablations confirming each component's contribution. Code is available at https://github.com/psaltaath/pbr-reid.