Geometrical shapes datasets
收藏资源简介:
Labeled datasets of images of geometrical shapes. Can be used for training and conceptual learning of supervised and unsupervised models. Use pickle, zip to extract raw images and labels. bw_shapes_3_1800 grayscale, 64x64, 3 distinct types (concepts), 600 per type 1800 total Grayscale circles, variation in shade, size, background Grayscale triangles, variation shade, size, background Empty (grey) backgrounds, variation in shade. Labels: 0 - 2 (circle, triangle, background) col_shapes_9_7200 color, 64x64, 9 distinct types (concepts), 800 per type 7200 total: - Two color (red, blue) circles, grey background, variation in shade, size, background - R,b triangles, grey background, variation shade, size, background - R,b horizontal stripes wide (red), narrow (blue) variation shade, width (r > 0.5 size, b < 0.5 size), background - R,b vertical stripes narrow (red), wide (blue) variation shade, width (r < 0.5 size, b > 0.5 size), background - Empty (grey) backgrounds, variation in shade. Labels 0,1 (circle r, b) 2,3 (triangle r, b) 4,5 (horizontal r,b) 6,7 (vertical r,b) 8 (background) Use pickle, zip to extract raw images and labels.
本数据集为带标注的几何形状图像数据集,可用于监督与无监督模型的训练及概念学习。可通过pickle与zip格式解压获取原始图像与标注标签。 本数据集包含两个子数据集: 1. bw_shapes_3_1800:该子数据集为灰度图像,分辨率为64×64,包含3种不同类别(概念),每类含600张样本,总计1800张。样本涵盖灰度圆形(亮度、尺寸与背景均存在变化)、灰度三角形(亮度、尺寸与背景均存在变化)以及灰度纯色背景(背景亮度可变)。其标注标签为0至2,分别对应圆形、三角形与背景。 2. col_shapes_9_7200:该子数据集为彩色图像,分辨率为64×64,包含9种不同类别(概念),每类含800张样本,总计7200张。样本涵盖以下类别: - 红蓝双色圆形,搭配灰色背景,亮度、尺寸与背景均存在变化; - 红蓝双色三角形,搭配灰色背景,亮度、尺寸与背景均存在变化; - 红蓝宽水平条纹(红色条带宽度大于图像尺寸的0.5,蓝色条带宽度小于0.5),亮度、条宽与背景均存在变化; - 红蓝窄垂直条纹(红色条带宽度小于图像尺寸的0.5,蓝色条带宽度大于0.5),亮度、条宽与背景均存在变化; - 灰度纯色背景,背景亮度可变。 其标注标签对应关系为:0、1分别对应红蓝双色圆形,2、3分别对应红蓝双色三角形,4、5分别对应红蓝水平条纹,6、7分别对应红蓝垂直条纹,8对应背景。




