Person Re-Identification by Camera Correlation Aware Feature Augmentation (1703.08837v1)

Abstract: The challenge of person re-identification (re-id) is to match individual images of the same person captured by different non-overlapping camera views against significant and unknown cross-view feature distortion. While a large number of distance metric/subspace learning models have been developed for re-id, the cross-view transformations they learned are view-generic and thus potentially less effective in quantifying the feature distortion inherent to each camera view. Learning view-specific feature transformations for re-id (i.e., view-specific re-id), an under-studied approach, becomes an alternative resort for this problem. In this work, we formulate a novel view-specific person re-identification framework from the feature augmentation point of view, called Camera coRrelation Aware Feature augmenTation (CRAFT). Specifically, CRAFT performs cross-view adaptation by automatically measuring camera correlation from cross-view visual data distribution and adaptively conducting feature augmentation to transform the original features into a new adaptive space. Through our augmentation framework, view-generic learning algorithms can be readily generalized to learn and optimize view-specific sub-models whilst simultaneously modelling view-generic discrimination information. Therefore, our framework not only inherits the strength of view-generic model learning but also provides an effective way to take into account view specific characteristics. Our CRAFT framework can be extended to jointly learn view-specific feature transformations for person re-id across a large network with more than two cameras, a largely under-investigated but realistic re-id setting. Additionally, we present a domain-generic deep person appearance representation which is designed particularly to be towards view invariant for facilitating cross-view adaptation by CRAFT.

• The paper proposes CRAFT, a novel framework that uses camera correlation aware feature augmentation to address feature distortions and improve person re-identification across camera views.
• CRAFT incorporates the underlying correlation between cameras using a parameterized feature augmentation mechanism, leading to more robust model learning beyond traditional methods.
• Empirical evaluation shows CRAFT consistently outperforms existing state-of-the-art methods on various datasets, achieving significant accuracy gains in multi-camera networks.

An Analysis of Camera Correlation Aware Feature Augmentation for Person Re-Identification

The paper "Person Re-Identification by Camera Correlation Aware Feature Augmentation" presents a novel approach to addressing the challenge of person re-identification (re-id) across non-overlapping camera views. Existing re-id techniques predominantly rely on learning view-generic transformations, potentially limiting their effectiveness in handling the distinct feature distortions introduced by varying camera perspectives. The authors propose an innovative view-specific framework named Camera coRrelation Aware Feature augmenTation (CRAFT) to counter these limitations, enabling more accurate person re-id.

Key Contributions

1. Novel Feature Augmentation Framework: The CRAFT algorithm introduces a unique perspective by modeling camera-specific feature augmentations. It employs a correlation-aware feature augmentation strategy, adapting original features into a new space that accounts for cross-view discrepancies. This adaptive space supports the learning of both view-specific and view-generic models, enhancing discrimination capabilities and mitigating feature distortion effects across camera views.
2. Incorporation of Camera Correlation: CRAFT utilizes the underlying correlation between cameras to quantify the shared information between them. This is achieved through a parameterized feature augmentation mechanism, moving beyond traditional zero-padding techniques. By incorporating the commonality between camera views, the algorithm ensures more robust model learning.
3. Scalability and Extension to Multi-Camera Networks: CRAFT is inherently scalable, allowing for collective re-id across large camera networks involving more than two cameras. This aspect extends typical pairwise camera analysis to more sophisticated and realistic network scenarios, signifying a practical advancement in re-id research.
4. Deep Representation for Person Appearance: The paper proposes a domain-generic deep learning approach for feature extraction, harnessing extensive auxiliary datasets. This methodology produces features that inherently possess greater invariance to view changes, bolstering the cross-view adaptability of re-id algorithms.

Evaluation and Results

The empirical analysis conducted across various challenging datasets including VIPeR, CUHK01, CUHK03, Market-1501, and QMUL GRID, demonstrates the effectiveness of the CRAFT model. The method consistently outperforms existing state-of-the-art algorithms, showcasing notable accuracy improvements. For instance, on VIPeR, a considerable rank-1 accuracy of 50.3% is achieved, surmounting prior leading methods by a substantive margin. Similar gains are observed across other datasets, affirming the broad applicability and superiority of the CRAFT approach.

Implications and Future Directions

The advancements introduced by the CRAFT framework signify a significant leap in the pursuit of accurate, scalable, and efficient person re-id mechanisms. By facilitating seamless integration with existing view-generic models, this approach demonstrates versatile adaptability and robustness in complex surveillance environments.

Future explorations could explore optimizing feature augmentation parameters for diverse camera characteristics and enhance the learning of non-linear transformations through more advanced neural architectures. Furthermore, extending the methodology to dynamic and temporal aspects of surveillance footage could potentially magnify its utility and effectiveness.

In sum, the CRAFT framework presents a significant contribution to the field of re-id by systematically leveraging camera correlations to augment features, setting a promising pathway toward more reliable and precise person identification systems in multi-camera networks.