ToolFG: Towards Well-Grounded Fine-Grained Image Classification

Fine-grained image classification (FGIC) has broad applications and has attracted significant research attention. In this paper, we explore a novel paradigm for solving FGIC by proposing $\mathbf{\text{ToolFG}}$, the first tool-integrated MLLM-based framework tailored to FGIC. ToolFG enables MLLMs to autonomously and flexibly use external tools during the reasoning process, actively interact with images, and collect verifiable visual cues for distinguishing highly similar categories in a more $\mathit{\text{reliable}}$ and $\mathit{\text{well-grounded}}$ manner. To equip the model with such tool-use ability, we design a novel $\mathbf{\text{MCTS-guided}}\mathbf{\text{tool-use}}\mathbf{\text{knowledge}}\mathbf{\text{distillation}}\mathbf{\text{mechanism}}$, which effectively mines tool-use- and FGIC-relevant knowledge from advanced proprietary MLLMs for model training. Furthermore, we propose a $\mathbf{\text{model-tool}}\mathbf{\text{co-evolution}}\mathbf{\text{mechanism}}$ that jointly refines the toolset and the model’s tool-use policy, driving them toward a mutually adapted and FGIC-specialized state. Extensive experiments demonstrate the effectiveness of our framework.