dappu97/Coil100-Augmented

该数据集源自Coil100，包含超过1.1百万张100个物体的图像。每个物体通过旋转台旋转360度，以不同的姿态相对于固定彩色相机拍摄，每5度拍摄一张图像，每个物体有72个姿态。此外，数据集还应用了平面旋转（9个角度）和18个缩放因子，以确保因素之间无依赖关系。数据集特别设计用于真实图像的分离表示学习，提供了物体、3D姿态、旋转和缩放等变化因素。

This dataset derives from Coil100, containing over 1.1 million images of 100 objects. Each object was photographed at 72 poses by rotating on a turntable at 5-degree intervals. Additionally, planar rotations (9 angles) and 18 scaling factors were applied to eliminate dependencies between factors. The objects have complex geometric and reflectance characteristics. This augmented version is specifically designed for Disentangled Representation Learning for real images. A binarized version of the dataset is also available.