ramya-ml/pick-a-filter-embeds

Pick-a-Filter是一个基于Pick-a-Pic v1构建的半合成数据集。数据集通过将用户随机分为偏好蓝色图像色调和偏好红色图像色调的两组来构建。对图像应用特定的赢和输过滤器，并根据混合比β随机选择样本添加过滤器。数据集包含了不同混合比下的版本，以及所有版本的OpenCLIP H/14嵌入。

Pick-a-Filter is a semi-synthetic dataset based on the Pick-a-Pic v1. It is constructed by dividing users randomly into two groups: those who prefer blue image tones and those who prefer red image tones. Specific winning and losing filters are applied to the images, and a certain percentage of samples are randomly selected to add filters according to the mixture ratio β. The dataset includes versions with different mixture ratios, as well as OpenCLIP H/14 embeddings for all versions.