This is a galaxy classification dataset from our work in the paper "Astronomical Image Superresolution Reconstruction with deep learning for better identifying interacting galaxies."
We obtained 1,302 clean samples, including 267 images of ring galaxies, 78 images of fan galaxies, 397 images of distorted galaxies, and 560 images of tail galaxies. We split the data into 64 training set, 16 validation set, and 20 testing set.
Table 1 gives the number of galaxy images in each class in each set. 
The galaxy dataset used in this study is based on the GALAXY CRUISE 1 (GC1) source catalog. GC DR1
GC DR1 is the first Japanese community science project based on data from the HSC-SSP in astronomy. GC DR1 have collected over two million independent classifications for 20,686 galaxies at z < 0.2. Among them, 20,116 galaxies are in the Wide layer, and 1,410 galaxies in the Deep/Ultra-Deep (D/UD) layer. 840 galaxies were classified twice: once at D/UD depth and once at Wide depth.