ACNet: Approaching-and-Centralizing Network for Zero-Shot Sketch-Based Image Retrieval

TL;DR

ACNet introduces a joint sketch-to-photo synthesis and retrieval framework that generates diverse photo-like images guided by retrieval to improve zero-shot sketch-based image retrieval, achieving state-of-the-art results.

Contribution

The paper proposes ACNet, a novel approach that jointly optimizes sketch synthesis and retrieval, effectively bridging domain and knowledge gaps for zero-shot SBIR.

Findings

Achieves state-of-the-art performance on ZS-SBIR datasets.

Effectively alleviates overfitting with diverse generated images.

Proxy-based NormSoftmax loss stabilizes training and enhances generalization.

Abstract

The huge domain gap between sketches and photos and the highly abstract sketch representations pose challenges for sketch-based image retrieval (\underline{SBIR}). The zero-shot sketch-based image retrieval (\underline{ZS-SBIR}) is more generic and practical but poses an even greater challenge because of the additional knowledge gap between the seen and unseen categories. To simultaneously mitigate both gaps, we propose an \textbf{A}pproaching-and-\textbf{C}entralizing \textbf{Net}work (termed "\textbf{ACNet}") to jointly optimize sketch-to-photo synthesis and the image retrieval. The retrieval module guides the synthesis module to generate large amounts of diverse photo-like images which gradually approach the photo domain, and thus better serve the retrieval module than ever to learn domain-agnostic representations and category-agnostic common knowledge for generalizing to unseen categories. These diverse images generated with retrieval guidance can effectively alleviate the overfitting problem troubling concrete category-specific training samples with high gradients. We also discover the use of proxy-based NormSoftmax loss is effective in the zero-shot setting because its centralizing effect can stabilize our joint training and promote the generalization ability to unseen categories. Our approach is simple yet effective, which achieves state-of-the-art performance on two widely used ZS-SBIR datasets and surpasses previous methods by a large margin.