Abstract
Despite the promising progress made in recent years, person re-identification (re-ID) remains a challenging task due to the complex variations in human appearances from different camera views. For this challenging problem, a large variety of algorithms have been developed in the fully-supervised setting, requiring access to a large amount of labeled training data. However, the main bottleneck for fully-supervised re-ID is the limited availability of labeled training samples. To address this problem, in this paper, we propose a self-trained subspace learning paradigm for person re-ID which effectively utilizes both labeled and unlabeled data to learn a discriminative subspace where person images across disjoint camera views can be easily matched. The proposed approach first constructs pseudo pairwise relationships among unlabeled persons using the k-nearest neighbors algorithm. Then, with the pseudo pairwise relationships, the unlabeled samples can be easily combined with the labeled samples to learn a discriminative projection by solving an eigenvalue problem. In addition, we refine the pseudo pairwise relationships iteratively, which further improves the learning performance. A multi-kernel embedding strategy is also incorporated into the proposed approach to cope with the non-linearity in person's appearance and explore the complementation of multiple kernels. In this way, the performance of person re-ID can be greatly enhanced when training data are insufficient. Experimental results on six widely-used datasets demonstrate the effectiveness of our approach and its performance can be comparable to the reported results of most state-of-the-art fully-supervised methods while using much fewer labeled data.
Introduction
Person re-identification (re-ID)—the task of matching an individual's identity across different, non-overlapping camera views—is notoriously difficult. Human appearance changes drastically due to shifts in lighting, pose, and camera angles. While deep learning and metric learning have made strides, they typically rely on massive amounts of labeled training data. In real-world scenarios, labeling thousands of image pairs for every new camera installation is impractical and expensive. When labeled data is scarce, supervised models tend to overfit, failing to generalize to new environments. This paper addresses the "labeling bottleneck" by proposing a framework that leverages unlabeled data to enhance discriminative learning.
The MKSSL Paradigm
The authors propose a Multi-Kernel Self-trained Subspace Learning (MKSSL) approach. The core idea is to learn a projection matrix that maps high-dimensional image features into a low-dimensional "common subspace" where images of the same person are close together.
The workflow follows a self-training loop:
- Initial Projection: A starting projection matrix is learned using the small amount of available labeled data.
- Subspace Projection: All unlabeled images are projected into this initial subspace. Because even a weak projection is better than raw features, the distance between images in this subspace is more meaningful.
- Pseudo-Relationship Building: A K-Nearest Neighbors (KNN) algorithm is used in the subspace to identify "pseudo-pairs"—unlabeled images that likely belong to the same person.
- Optimization: True labels and pseudo-labels are combined into a single objective function using a graph Laplacian regularization term, solved as an eigenvalue problem to find a superior projection.
- Iterative Refinement: The process repeats. With each iteration, the subspace becomes more discriminative, the pseudo-labels become more accurate, and the projection improves.
To handle the non-linearity of human appearance, the framework incorporates Multi-Kernel Embedding, which fuses multiple feature representations (like the GOG descriptor) and kernel types into the learning process.
Experimental Evaluation
The researchers evaluated MKSSL on six widely-used person re-ID benchmarks: VIPeR, CUHK01, PRID2011, PRID450S, GRID, and 3DPeS. These datasets represent a variety of challenges, including significant viewpoint changes, low resolution, and heavy background clutter.
Key Results and Insights
The MKSSL method consistently outperformed other semi-supervised baselines and rivaled fully supervised methods. A critical insight is that constructing a KNN graph on raw visual features is often misleading due to camera noise; by moving to a learned subspace first, the neighborhood structure becomes cleaner, making pseudo-labels reliable for training.
As shown in the Cumulative Matching Characteristics (CMC) curves, MKSSL demonstrates high label efficiency. On the VIPeR dataset with only 1/3 of the data labeled, MKSSL achieved a 40.6% Rank-1 accuracy, a significant jump over the fully supervised baseline (MKFSL), which managed only 28.9% with the same limited data. Furthermore, the semi-supervised MKSSL often performed nearly as well as established fully supervised methods like XQDA that utilized the full training set.
Conclusion
MKSSL effectively bridges the gap between limited labeled data and abundant unlabeled images. While the method requires a minimum initialization threshold (typically a label ratio of at least 1/20) to avoid model drift, it offers a robust solution for real-world surveillance where labeling is a bottleneck. Future work may explore integrating this self-training loop directly into deep neural network architectures for end-to-end semi-supervised re-ID.
Summary updated