HC-M3D
收藏资源简介:
HC-M3D数据集是由华东师范大学计算机科学与技术学院的研究人员创建的一种新型多模态数学推理数据集。该数据集包含1851个样本,这些样本都是经过人工精心挑选的,确保了问题解答对图像的依赖性。HC-M3D的特别之处在于,它为超过400个问题提供了相似但不同的图像,这些图像的改变会导致正确答案的变化。这个数据集的目的是为了测试模型是否能够识别这些细微的视觉差异,并据此做出正确的预测。
The HC-M3D dataset is a novel multimodal mathematical reasoning dataset created by researchers from the School of Computer Science and Technology, East China Normal University. It contains 1,851 samples, all manually curated to ensure that problem-solving relies on the accompanying images. What sets HC-M3D apart is that it provides similar yet distinct images for over 400 problems, where alterations to these images will lead to changes in the correct answers. The core objective of this dataset is to test whether models can recognize these subtle visual differences and make accurate predictions accordingly.

- 1The Role of Visual Modality in Multimodal Mathematical Reasoning: Challenges and Insights华东师范大学计算机科学与技术学院 · 2025年



