RC-49

RC-49 是一个基准数据集，用于生成以连续标量变量为条件的图像。它是通过单独渲染来自 ShapeNet 的 49 个 3-D 椅子模型制成的。每个椅子模型以 899 个偏航角从 $0.1^{\circ}$ 到 $89.9^{\circ}$ 渲染，步长为 $0.1^{\circ}$。该数据集包含 44,051 个大小为 $64\times64$ 的 RGB 图像，并带有相应的偏航角作为标签。请注意，在 CcGAN 中，如果最后一位数字是奇数，则使用角度进行训练。因此，训练集中有 450 个角度。此外，对于这 450 个训练角度，每个角度仅使用 25 张图像进行训练。

RC-49 is a benchmark dataset for generating images conditioned on continuous scalar variables. It is constructed by independently rendering 49 3-D chair models sourced from ShapeNet. Each chair model is rendered across 899 yaw angles ranging from 0.1° to 89.9° with a step size of 0.1°. This dataset contains 44,051 RGB images with a resolution of 64×64, paired with their corresponding yaw angles as labels. Please note that in CcGAN, angles are used for training if their last digit is odd. As a result, there are 450 angles in the training set. Additionally, for these 450 training angles, only 25 images are utilized per angle during training.

搜集汇总

背景与挑战

背景概述

RC-49是一个包含44,051张64x64 RGB图像的基准数据集，用于生成以连续偏航角为条件的图像，每张图像带有对应的偏航角标签。数据集通过渲染49个3-D椅子模型制成，训练集使用了450个角度，每个角度包含25张图像。