Toward Adaptive Person Search/ReID systems

Document Type

Dissertation

Abstract

Person re-identification (ReID) is a challenging computer vision problem where the objective is to retrieve a person of interest from a gallery of images. On the other hand, Person Search (PS) tries to localize and ReID persons in scenes. The problem itself is a critical challenge in the area of smart city security applications. The motivation is to solve the issue of large-scale variation, occlusion, pose, and viewpoint variations. Conventional person ReID methods struggle to generalize across different domains, leading to inferior cross-domain performance. To address this, domain generalization (DG) person Reid methods are proposed. The majority of DG studies focus on designing more robust models that are trained on source datasets in an offline setting with explicit components or training mechanisms for domain generalization. Moreover, these frameworks do not take into consideration that in the real world, the environment is in continual change resulting in the degradation of the discriminative feature extraction. To tackle these problems, we propose a new problem formulation of Test-time adaption for ReID (TTA ReID). Our TTA ReID takes a pre-trained model that was not designed for domain generalization from the source domain and aims in inference to adapt it to the current target domain in an unsupervised setting.

Publication Date

6-2023

Comments

Thesis submitted to the Deanship of Graduate and Postdoctoral Studies

In partial fulfillment of the requirements for the M.Sc degree in Computer Vision

Advisors: Dr. Hisham Cholakkal, Dr. Fahad Khan

with 2 year embargo period

This document is currently not available here.

Share

COinS