Smart Closet: Recommending outfits from your closet

figure

Our work is recommending outfits from fashion items. The challenges are that the features of items are complex and difficult to obtain, and the exponential combinatorial complexity. We use a neural network that learn from large dataset to extract image features, and use beam search to reduce the computational complexity. We also create a large dataset from a popular fashion sharing website for both training and evaluation. We compare our system with randomly created outfits. According to our outfit recommending evaluation, our recommender can create good-looking outfit from a pool of apparently random items significantly better than random.

Download