New benchmark reveals MLLM struggles with long-term temporal understanding in remote sensing

Researchers have introduced ChronoBench, a new benchmark designed to evaluate the long-term temporal understanding capabilities of multimodal large language models (MLLMs) in remote sensing. The benchmark revealed that current MLLMs significantly underperform human experts, particularly in long-term memory tasks. To address this, the team developed GeoChrono, an MLLM incorporating a Temporal Trajectory Encoder and a Coarse-to-Fine Token Compressor to improve tracking, memorization, and reasoning about geographic evolution, achieving state-of-the-art results on ChronoBench. AI

IMPACT This research highlights critical limitations in current MLLMs for long-term temporal reasoning, potentially guiding future model development for applications requiring historical context.

RANK_REASON The cluster describes a new academic paper introducing a benchmark and a new model for evaluating temporal understanding in remote sensing. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• C2FComp
• ChronoBench
• ChronoInstruct
• DagsHub
• Hugging Face
• TempEnc

• paper
• model release
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →