Search papers, labs, and topics across Lattice.
This paper introduces PailitaoGR, a Latent Think-with-Images method designed to enhance generative image retrieval by focusing on target identification and selective use of auxiliary evidence. The approach integrates a target-focused perception mechanism and a selective auxiliary-evidence utilization mechanism, allowing the model to effectively zoom in on relevant content without cropping and read contextual information without relying on OCR. Experimental results demonstrate that PailitaoGR outperforms existing methods by an average of 13.8%, highlighting its potential for improving image search accuracy in real-world applications.
Generative image retrieval just got a major upgrade鈥擯ailitaoGR boosts performance by 13.8% by mastering target focus and auxiliary evidence utilization.
Generative retrieval has demonstrated strong performance by directly generating product semantic identifiers (SIDs). Extending this paradigm to image search, however, is nontrivial because real-world query images contain diverse information, including the search target, useful auxiliary evidence, and irrelevant visual content. This requires the model to identify and focus on the search target while selectively utilizing auxiliary evidence. In this paper, we propose \textbf{PailitaoGR}, a \emph{Latent Think-with-Images} method for generative image retrieval, which internalizes target-focused perception and selective auxiliary-evidence utilization into a the generative retrieval model, enabling \textit{Zooming without Cropping} and \textit{Reading without OCR}. Specifically, we design a target-focused perception mechanism that identifies and enhances visual tokens of the search target, consisting of a target Enhancer and a learning strategy based on on-policy distillation and attention guidance loss, enabling the model to focus on search-target regions. We also design a selective auxiliary-evidence utilization mechanism that identifies and enhances visual tokens of auxiliary evidence, including an auxiliary enhancer and an in-capacity incremental contrastive distillation strategy, enabling the model to exploit auxiliary evidence. We construct training and validation sets sampled from real-world online image-search logs. Experiments show that our method outperforms existing baselines by an average of 13.8\%, validating its effectiveness.
