Fashion-ZSD
收藏资源简介:
Fashion-ZSD是一个为评估零样本检测方法而生成的玩具数据集,基于Fashion-MNIST数据集。该数据集包含70,000个28x28大小的灰度图像,分为10个类别。数据集被分为训练集和测试集,其中训练集包含7个类别,测试集包含3个类别。每个图像包含三个不同的对象,考虑了四种场景:无噪声或遮挡、部分遮挡、杂乱和既有部分遮挡又有杂乱的场景。数据集包括8333个训练图像、8000个验证图像和6999个测试图像。
Fashion-ZSD is a toy dataset generated for evaluating zero-shot detection methods, based on the Fashion-MNIST dataset. It comprises 70,000 grayscale images of size 28x28, categorized into 10 classes. The dataset is divided into a training set and a test set, with the training set containing 7 classes and the test set containing 3 classes. Each image includes three distinct objects, considering four scenarios: no noise or occlusion, partial occlusion, clutter, and scenarios with both partial occlusion and clutter. The dataset includes 8,333 training images, 8,000 validation images, and 6,999 test images.
数据集概述
名称: Fashion-ZSD
来源: 基于Fashion-MNIST数据集生成,用于评估Zero-Shot Detection(ZSD)方法。
数据集大小:
- 训练集: 8333张图像
- 验证集: 8000张图像
- 测试集: 6999张图像
图像特征:
- 图像为28x28像素的灰度图像
- 每张图像包含三个不同的对象
- 包含四种场景:无噪声或遮挡、部分遮挡、杂乱场景、以及既有部分遮挡又有杂乱区域的场景
类别:
- 训练类别: tshirt, trouser, coat, sandal, shirt, sneaker, bag
- 测试类别: ankle_boot, dress, pullover
数据集内容:
- 训练、验证和测试分割的图像数据
- 训练和测试类别的词嵌入向量
- 评估脚本
引用信息:
@inproceedings{Demirel2018ZeroShotOD, title={Zero-Shot Object Detection by Hybrid Region Embedding}, author = {Demirel, Berkan and Cinbis, Ramazan Gokberk and Ikizler-Cinbis, Nazli}, booktitle={BMVC}, year={2018} }




