Vector of Locally and Adaptively Aggregated Descriptors for Image Feature Representation. (August 2021)

• Record Type: Journal Article
• Authors: Zhang, Jian; Cao, Yunyin; Wu, Qun
• Is Part Of: Pattern recognition. Volume 116(2021)
• Publication Date: 2021-08
• Subjects: VLAD -- Deep learning -- Weighting scheme -- Gating scheme -- Feature representation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.107952
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: VLAAD framework uses adaptive weighting mechanism to improve the discriminative power of feature representations. VLAAD framework is implemented by the Weighted NetVLAD and Gated NetVLAD. Competitive performance has been achieved comparing with the state-of-the-art. Abstract: VLAD (Vector of Locally Aggregated Descriptors) has been widely adopted in image representation. However, the VLAD algorithm seeks for the algebraic sum of the residue vectors between the descriptors and the centroid of cluster they belong to, and this could decrease the discriminative power of feature representations. To this end, this paper originally proposes a VLAAD (Vector of Locally and Adaptively Aggregated Descriptors) framework to adaptively assign a weight to each residue vector. First, we compute the weights using the magnitude of each residue vector, and encapsulate the weighted VLAD block into ResNet to form an end-to-end Weighted NetVLAD method. To further enhance the discriminative power of the features, we subsequently replace the magnitude-based weight computation with a gating scheme to achieve automatic weight estimation. The enhanced version is named as Gated NetVLAD method. The experimental results on CIFAR-10, MNIST Digits, Pittsburgh Google street view and ImageNet-Dog datasets demonstrate the promotion in classification accuracy and retrieval mAP using VLAAD against several state-of-the-art methods.