Calibrated Domain-Invariant Learning for Highly Generalizable Large Scale Re-Identification

TL;DR

This paper introduces ADIN, a novel adversarial learning framework that enhances large-scale re-identification by explicitly disentangling identity features from variations, utilizing free annotations and a calibrated loss to improve cross-dataset generalization.

Contribution

The paper proposes a new adversarial domain-invariant learning method for large-scale ReID that leverages free annotations and a calibrated loss to improve generalization across datasets.

Findings

ADIN outperforms existing methods in cross-dataset transfer tasks.

Utilizing free annotations improves feature disentanglement.

Calibrated adversarial loss mitigates nuisance class imbalance.

Abstract

Many real-world applications, such as city-scale traffic monitoring and control, requires large-scale re-identification. However, previous ReID methods often failed to address two limitations in existing ReID benchmarks, i.e., low spatiotemporal coverage and sample imbalance. Notwithstanding their demonstrated success in every single benchmark, they have difficulties in generalizing to unseen environments. As a result, these methods are less applicable in a large-scale setting due to poor generalization. In seek for a highly generalizable large-scale ReID method, we present an adversarial domain invariant feature learning framework (ADIN) that explicitly learns to separate identity-related features from challenging variations, where for the first time "free" annotations in ReID data such as video timestamp and camera index are utilized. Furthermore, we find that the imbalance of nuisance classes jeopardizes the adversarial training, and for mitigation we propose a calibrated adversarial loss that is attentive to nuisance distribution. Experiments on existing large-scale person vehicle ReID datasets demonstrate that ADIN learns more robust and generalizable representations, as evidenced by its outstanding direct transfer performance across datasets, which is a criterion that can better measure the generalizability of large-scale ReID methods/