Zero Shot Sketch Based Image

Zero-shot sketch-based image retrieval (ZS-SBIR) aims to find images matching a hand-drawn sketch, even without training examples of that specific sketch-image pair. Current research heavily utilizes large pre-trained vision-language models and transformers, often incorporating multi-modal prompt learning, cross-modal attention mechanisms, and novel loss functions to bridge the significant semantic gap between sketches and photos. These advancements focus on improving both the accuracy and explainability of the retrieval process, addressing challenges like domain adaptation and fine-grained classification. The success of ZS-SBIR has implications for various applications, including content-based image search, design assistance, and forensic science.