GeoDCL: Weak Geometrical Distortion based Contrastive Learning for Fine-grained Fashion Image Retrieval | IEEE Journals & Magazine | IEEE Xplore

GeoDCL: Weak Geometrical Distortion based Contrastive Learning for Fine-grained Fashion Image Retrieval


Impact Statement:Fine-grained fashion image retrieval (FIR) is a critical research task in fashion artificial intelligence (AI). However, current methods often overlook the importance of ...Show More

Abstract:

This paper addresses fine-grained fashion image retrieval (FIR), which aims at the detailed and precise retrieval of fashion items from extensive databases. Conventional ...Show More
Impact Statement:
Fine-grained fashion image retrieval (FIR) is a critical research task in fashion artificial intelligence (AI). However, current methods often overlook the importance of multi-view fashion images. This paper proposes a GeoDCL strategy that can be seamlessly integrated into existing fine-grained FIR methods, significantly enhancing retrieval accuracy without increasing inference time. This approach goes beyond the simple application of contrastive learning to fine-grained FIR by incorporating carefully designed positive pairs, including geometric distortion scales, types, numbers, and contrastive loss. In addition, we provide an in-depth analysis of these designs. These innovations in fine-grained FIR offer valuable references for researchers working in fashion AI.

Abstract:

This paper addresses fine-grained fashion image retrieval (FIR), which aims at the detailed and precise retrieval of fashion items from extensive databases. Conventional fine-grained FIR methods design complex attention modules to enhance attribute-aware feature discrimination. However, they often ignore the multi-view characteristics of real-world fashion data, leading to diminished model accuracy. Furthermore, our empirical analysis revealed that the straightforward application of standard contrastive learning methods to fine-grained FIR often yields suboptimal results. To alleviate this issue, we propose a novel weak geometrical distortion-based contrastive learning (GeoDCL) strategy. Specifically, GeoDCL incorporates both a novel positive pair design and a novel contrastive loss. GeoDCL can be seamlessly integrated into state-of-the-art (SOTA) fine-grained FIR methods during the training stage to enhance performance during inference. When GeoDCL is applied, the model structures of ...
Published in: IEEE Transactions on Artificial Intelligence ( Early Access )
Page(s): 1 - 13
Date of Publication: 28 February 2025
Electronic ISSN: 2691-4581

Contact IEEE to Subscribe