Transformers and CNNs both Beat Humans on SBIR

TL;DR

This paper demonstrates that vision transformers significantly outperform CNNs and humans in sketch-based image retrieval, addressing flip invariance issues and proposing improved models for the SBIR task.

Contribution

It introduces modifications for better flip equivariance in SBIR models and shows that vision transformers outperform CNNs and humans on large-scale benchmarks.

Findings

Vision transformers outperform CNNs in SBIR.

Proposed modifications improve flip invariance.

Achieved state-of-the-art recall of 62.25% on Sketchy.

Abstract

Sketch-based image retrieval (SBIR) is the task of retrieving natural images (photos) that match the semantics and the spatial configuration of hand-drawn sketch queries. The universality of sketches extends the scope of possible applications and increases the demand for efficient SBIR solutions. In this paper, we study classic triplet-based SBIR solutions and show that a persistent invariance to horizontal flip (even after model finetuning) is harming performance. To overcome this limitation, we propose several approaches and evaluate in depth each of them to check their effectiveness. Our main contributions are twofold: We propose and evaluate several intuitive modifications to build SBIR solutions with better flip equivariance. We show that vision transformers are more suited for the SBIR task, and that they outperform CNNs with a large margin. We carried out numerous experiments and introduce the first models to outperform human performance on a large-scale SBIR benchmark (Sketchy). Our best model achieves a recall of 62.25% (at k = 1) on the sketchy benchmark compared to previous state-of-the-art methods 46.2%.