Work from a few months ago on using Densenets for galaxy classification, and also using pseudo-labelling for training. Performance is mediocre (didn't experiment with training hyperparameters much), getting 0.08438 on public benchmark.
Kaggle competition link: https://www.kaggle.com/c/galaxy-zoo-the-galaxy-challenge
DenseNet paper: https://arxiv.org/abs/1608.06993
Pseudo-Label paper: http://deeplearning.net/wp-content/uploads/2013/03/pseudo_label_final.pdf