追星星的AI绘本测试集

license: Apache License 2.0
---

本数据集为 AI+公益大赛系列 “追星星的AI”·孤独症儿童绘本LoRA训练挑战 赛题指定测试数据。

本测试集包括 常识认知、社交礼仪、心智解读、趣味故事 共4个类型各5个故事，即共20个故事，每个故事包括7-10句不等，每句话为一幕故事内容，如：

心智解读-故事1：《小朋友不借给我玩具》
第1幕：乐乐和朵朵在一起玩。
第2幕：乐乐很喜欢朵朵的玩具小车。
第3幕：乐乐走到朵朵身边，问："这个小车借我玩，好不好？"
第4幕：朵朵摇摇头说："不好，我现在还想玩。"
第5幕：因为朵朵不借小车，乐乐心里有点难过。
第6幕：乐乐想：我可以等一会儿再问问朵朵。
第7幕：乐乐看到旁边有积木，就开始搭积木。
第8幕：他搭了一个高高的城堡，觉得很好玩。
第9幕：乐乐知道了：小朋友不借玩具没关系，我可以先玩别的。

赛题要求:
评测LoRA效果需在4个类型中各选用至少2个故事文本进行图像推理创作。

#### 下载方法
:modelscope-code[]{type="sdk"}
:modelscope-code[]{type="git"}

License: Apache License 2.0
---
This dataset is the designated test data for the "AI for Public Welfare Competition Series: Star-Chasing AI · Autism Children's Picture Book LoRA Training Challenge" track.

This test set includes 5 stories for each of the 4 categories: common sense cognition, social etiquette, theory of mind, and interesting stories, totaling 20 stories. Each story consists of 7 to 10 sentences, with each sentence representing a scene of the story. For example:

Theory of Mind - Story 1: *The Kid Who Was Denied a Toy Borrow*
Scene 1: Lele and Duoduo are playing together.
Scene 2: Lele is very fond of Duoduo's toy car.
Scene 3: Lele walks up to Duoduo and asks, "Can I borrow this toy car to play with?"
Scene 4: Duoduo shakes her head and says, "No, I still want to play with it right now."
Scene 5: Lele feels a little sad because Duoduo didn't lend him the toy.
Scene 6: Lele thinks to himself: I can ask Duoduo again later.
Scene 7: Lele sees some building blocks nearby and starts building with them.
Scene 8: He builds a tall castle and finds it very enjoyable.
Scene 9: Lele learns: It's okay if a kid doesn't lend their toy, I can play with something else first.

Competition Track Requirements:
To evaluate the LoRA effect, at least 2 story texts from each of the 4 categories must be selected for image inference and creation.

#### Download Method
:modelscope-code[]{type="sdk"}
:modelscope-code[]{type="git"}

搜集汇总

背景与挑战

背景概述

该数据集是'AI+公益'系列比赛中'追星星的AI'自闭症儿童绘本LoRA训练挑战的专用测试集，包含常识认知、社交礼仪、心理理论和趣味故事四个类别，共20个故事，每个故事由7到10个句子组成。挑战要求从每个类别中至少选取2个故事进行图像推理创作，以评估LoRA训练效果。