#MsEarth
May 28, 2025 at 5:57 AM
enhance the development and evaluation of MLLMs in scientific reasoning. The benchmark is publicly available to foster further research and innovation in this field. Resources related to this benchmark can be found at https://huggingface.co/MSEarth [7/8 of https://arxiv.org/abs/2505.20740v1]
May 28, 2025 at 5:57 AM
MSEarth supports a variety of tasks, including scientific figure captioning, multiple choice questions, and open-ended reasoning challenges. By bridging the gap in graduate-level benchmarks, MSEarth provides a scalable and high-fidelity resource to [6/8 of https://arxiv.org/abs/2505.20740v1]
May 28, 2025 at 5:57 AM
benchmark curated from high-quality, open-access scientific publications. MSEarth encompasses the five major spheres of Earth science: atmosphere, cryosphere, hydrosphere, lithosphere, and biosphere, featuring over 7K figures with refined captions. [4/8 of https://arxiv.org/abs/2505.20740v1]
May 28, 2025 at 5:57 AM
datasets or simplistic figure-caption pairs, which do not adequately reflect the intricate reasoning and domain-specific insights required for real-world scientific applications. To address these gaps, we introduce MSEarth, a multimodal scientific [3/8 of https://arxiv.org/abs/2505.20740v1]
May 28, 2025 at 5:57 AM
Zhao, Xu, Liu, Zhou, Ling, Fei, Yue, Bai, Zhang, Wu: MSEarth: A Benchmark for Multimodal Scientific Comprehension of Earth Science https://arxiv.org/abs/2505.20740 https://arxiv.org/pdf/2505.20740 https://arxiv.org/html/2505.20740
May 28, 2025 at 5:57 AM