SARE: Adaptive Reasoning for Fine-Grained Visual Recognition

A new framework called SARE (Sample-wise Adaptive Reasoning) has been proposed for training-free Fine-Grained Visual Recognition (FGVR) using Large Vision-Language Models (LVLMs). The method addresses two key limitations of existing approaches: uniform inference across samples with varying difficulty, and lack of error experience reuse. SARE employs a cascaded design combining fast candidate retrieval with adaptive reasoning, improving accuracy and efficiency. The paper is available on arXiv under ID 2603.17729.