Recent advances in deep learning have shown their ability to learn strong\nfeature representations for images. The task of image clustering naturally\nrequires good feature representations to capture the distribution of the data\nand subsequently differentiate data points from one another. Often these two\naspects are dealt with independently and thus traditional feature learning\nalone does not suffice in partitioning the data meaningfully. Variational\nAutoencoders (VAEs) naturally lend themselves to learning data distributions in\na latent space. Since we wish to efficiently discriminate between different\nclusters in the data, we propose a method based on VAEs where we use a Gaussian\nMixture prior to help cluster the images accurately. We jointly learn the\nparameters of both the prior and the posterior distributions. Our method\nrepresents a true Gaussian Mixture VAE. This way, our method simultaneously\nlearns a prior that captures the latent distribution of the images and a\nposterior to help discriminate well between data points. We also propose a\nnovel reparametrization of the latent space consisting of a mixture of discrete\nand continuous variables. One key takeaway is that our method generalizes\nbetter across different datasets without using any pre-training or learnt\nmodels, unlike existing methods, allowing it to be trained from scratch in an\nend-to-end manner. We verify our efficacy and generalizability experimentally\nby achieving state-of-the-art results among unsupervised methods on a variety\nof datasets. To the best of our knowledge, we are the first to pursue image\nclustering using VAEs in a purely unsupervised manner on real image datasets.\n