multi-modal-llm

RS-Claw proposes an active tool exploration paradigm for remote sensing agents using hierarchical skill trees, enabling on-demand sequential decision-making and achieving up to 86% input token compression while outperforming passive selection baselines on Earth-Bench.

This paper introduces FoodCHA, a multi-modal LLM agent framework designed for fine-grained food analysis, addressing challenges in hierarchical consistency and attribute discrimination for dietary monitoring.