Scribble

Scribble 是一个新的大纲数据集，由 200 个图像（150 个火车，50 个测试）组成，每个类别有 10 个类别——篮球、鸡肉、饼干、纸杯蛋糕、月亮、橙子、足球、草莓、西瓜和菠萝。所有图像都有白色背景，并使用流行搜索引擎上的搜索关键字收集。在每幅图像中，我们获得了图像的粗略轮廓。在将图像阈值化为黑白图像后，我们找到了图像中最大的斑点。我们填充最大斑点的内部孔洞，并使用 SavitzkyGolay 过滤器获得平滑的轮廓。

Scribble is a novel outline dataset consisting of 200 images (150 for training, 50 for testing), covering 10 categories: basketball, chicken, biscuit, cupcake, moon, orange, football, strawberry, watermelon, and pineapple. All images have a white background and were collected using search keywords from popular search engines. For each image, the rough outline of the target object was obtained. After thresholding the image into a binary black-and-white version, the largest connected blob in the image was identified. We filled the internal holes of this largest blob, and applied the Savitzky-Golay filter to acquire a smooth contour.

搜集汇总

背景与挑战

背景概述

Scribble是一个用于草图到图像翻译的图像数据集，包含200张图像（150张训练、50张测试），涵盖10个类别。这些图像通过搜索关键词收集，并经过处理生成平滑的轮廓。