Multi-Scale Learning for Low-Resolution Person Re-Identification

Abstract

In real world person re-identification (re-id), images of people captured at very different resolutions from different locations need be matched. Existing re-id models typically normalise all person images to the same size. However, a low-resolution (LR) image contains much less information about a person, and direct image scaling and simple size normalisation as done in conventional re-id methods cannot compensate for the loss of information. To solve this LR person re-id problem, we propose a novel joint multi-scale learning framework, termed joint multi-scale discriminant component analysis (JUDEA). The key component of this framework is a heterogeneous class mean discrepancy (HCMD) criterion for cross-scale image domain alignment, which is optimised simultaneously with discriminant modelling across multiple scales in the joint learning framework. Our experiments show that the proposed JUDEA framework outperforms existing representative reid methods as well as other related LR visual matching models applied for the LR person re-id problem.

Deeply-Learned Part-Aligned Representations for Person Re-identification

“…A lot of similarity/metric learning techniques [57,58,33,27,19] have been applied or designed to learn metrics, robust to light/view/pose changes, for person matching. The recent developments include soft and probabilistic patch matching for handling pose misalignment [4,3,36], similarity learning for dealing with probe and gallery images with different resolutions [24,17], connection with transfer learning [34,38], reranking inspired by the connection with image search [65,13], partial person matching [66], human-in-the-loop learning [30,46], and so on.…”

Query-Guided End-To-End Person Search

“…Person Re-Identification. Classic approaches for person re-identification have focused on manual feature design [33,11,9,43] and metric learning [20,44,17,19,21,27,26]. As in object detection, CNNs have recently conquered the scene in re-identification, too [1,18,41].…”

Resource Aware Person Re-identification Across Multiple Resolutions

“…Some works explicitly consider local features or multiscale features in the neural networks [11,27,30,37,47,56,57]. By contrast, we implicitly combine features across scale and abstraction by tapping into the different stages of the convolutional network.…”