SHAPES (Swarm Heuristics based Adaptive and Penalized Estimation of Splines)

SHAPES 是一个合成图像数据集，旨在对系统进行基准测试，以了解多个对象之间的空间和逻辑关系。该数据集包含有关彩色形状排列的复杂问题。这些问题是围绕概念和关系的组合构建的，例如圆圈上方有红色形状吗？还是红色的形状是蓝色的？问题包含两到四个属性、对象类型或关系。总共有 244 个问题和 15,616 个图像，所有问题都有一个是和否的答案（以及相应的支持图像）。这消除了学习偏差的风险。
每个图像都是一个 30×30 RGB 图像，描绘了一个 3×3 网格对象。每个对象都以形状（圆形、方形、三角形）、颜色（红色、绿色、蓝色）和大小（小、大）为特征。

SHAPES is a synthetic image dataset designed to benchmark systems for understanding spatial and logical relationships between multiple objects. This dataset contains complex questions regarding the arrangement of colored shapes. These questions are constructed around combinations of concepts and relationships, such as "Is there a red shape above a circle?" or "Is a red shape blue?" Each question includes 2 to 4 attributes, object types, or relationships. There are a total of 244 questions and 15,616 images, with every question paired with both a "yes" and "no" answer (and their corresponding supporting images), which eliminates the risk of learning bias. Each image is a 30×30 RGB image depicting a 3×3 grid of objects. Each object is characterized by its shape (circle, square, triangle), color (red, green, blue), and size (small, large).

搜集汇总

背景与挑战

背景概述

SHAPES是一个合成图像数据集，旨在评估系统对多对象间空间与逻辑关系的理解能力。它包含244个问题和15,616张30×30 RGB图像，每个图像以3×3网格展示不同形状、颜色和大小的物体，所有问题均设计为是/否答案以避免学习偏差。