This dataset consists of pairs of anime images with differential changes, where images within the same pair exhibit highly similar features or composition. The dataset is used for training contrastive learning models, aiming to determine whether two images belong to the same differential set. The model plays a crucial role during the dataset filtering stage.

The data is crawled from Sankaku and before training, it is recommended to group the images within each set based on their aspect ratios. This grouping helps mitigate the potential impact of data impurities on the training process.

Dataset	Groups	Images	Description
v0	19898	63212	Native dataset crawled from sankaku