arielnlee/Realistic-Occlusion-Dataset
收藏资源简介:
真实遮挡数据集(ROD)是通过精心设计的对象收集协议创建的,旨在收集和捕捉来自16个类别的40多个不同对象的图像:香蕉、棒球、牛仔帽、杯子、哑铃、锤子、笔记本电脑、微波炉、鼠标、橙子、枕头、盘子、螺丝刀、煎锅、铲子和花瓶。图像在明亮的房间内使用iPhone 13 Pro超广角摄像头拍摄,背景为棕色木桌和纯色墙壁。遮挡物为红色或蓝色的木块或纸板,位置在相机与主物体之间,沿x轴从左到右变化。每个对象包括1张清晰图像和12张遮挡图像,总计1231个样本。该数据集用于测试模型在遮挡情况下的鲁棒性,特别是在论文《Hardwiring ViT Patch Selectivity into CNNs using Patch Mixing》中。
Real-world Occlusion Dataset (ROD) was created through a meticulously designed object collection protocol, aiming to collect and capture images of over 40 distinct objects across 16 categories: banana, baseball, cowboy hat, cup, dumbbell, hammer, laptop, microwave oven, mouse, orange, pillow, plate, screwdriver, frying pan, shovel, and vase. The images were captured in a bright room using the ultra-wide-angle camera of an iPhone 13 Pro, with a brown wooden table and solid-color walls as the background. The occluders are red or blue wooden blocks or cardboard pieces, positioned between the camera and the main object and varying from left to right along the x-axis. Each object includes 1 clear image and 12 occlusion images, totaling 1231 samples. This dataset is used to test the robustness of models under occlusion conditions, and was specifically applied in the paper titled "Hardwiring ViT Patch Selectivity into CNNs using Patch Mixing".
Real Occlusion Dataset (ROD) 数据集概述
数据集基本信息
- 许可证: other
- 任务类别: image-classification
- 语言: en
- 标签: occlusion
- 大小类别: 1K<n<10K
数据集特征
- 图像 (image): 数据类型为图像。
- 标签 (label): 数据类型为类别标签,包含以下类别名称:
- 0: banana
- 1: baseball
- 2: cowboy hat
- 3: cup
- 4: dumbbell
- 5: hammer
- 6: laptop
- 7: microwave
- 8: mouse
- 9: orange
- 10: pillow
- 11: plate
- 12: screwdriver
- 13: skillet
- 14: spatula
- 15: vase
数据集分割
- 名称: ROD
- 字节数: 3306212413
- 示例数量: 1231
- 下载大小: 3285137456
- 数据集大小: 3306212413
数据集描述
- 对象收集: 包含16个类别,超过40个不同对象。
- 拍摄环境: 在明亮的房间内,使用自然柔和光线,所有对象放置在棕色木桌上,背景为单色墙。
- 拍摄设备: iPhone 13 Pro 超广角相机,使用三脚架,拍摄角度约90度,距离对象1米。
- 遮挡物: 木块或纸板,涂成红色或蓝色,放置在相机与主对象之间,x轴位置变化,从画面左侧移动至右侧。
- 图像数量: 每个对象拍摄1张清晰图像和12张遮挡图像。




