遇见数据集

neoneye/simon-arc-solve-cross-v6

收藏
Hugging Face2024-11-23 更新2024-12-14 收录
官方服务:

资源简介:

simons ARC(抽象与推理语料库)解决交叉版本6数据集,任务类别包括图像到文本和文本到图像,语言为英语,数据规模在1K到10K之间。该数据集的主要任务是通过识别两条线的交叉点、最上面和最下面的线来进行推理。每个版本的更新主要集中在图像大小、任务类型和数据处理方式的变化上。

The Simons ARC dataset is a cross-version abstraction and reasoning corpus, primarily used for image-to-text and text-to-image tasks. The dataset includes multiple versions, each with different task types and image sizes. Version 1 tasks involve identifying the intersection point of two lines and determining which line is on top or bottom. Version 2 and Version 3 increased the image size, and Version 3 added a new task type, which is identifying the lines that pass through the intersection point. Version 4 added earlier predictions to some rows. Version 5 added new fields such as `arc_task`, `test_index`, and `earlier_output`. Version 6 replaced the RLE compressed response with a raw pixel response.

提供机构:
neoneye
二维码
社区交流群
二维码
科研交流群
商业服务