Fashion outfit completion using bidirectionnal LSTM.

Summary:

• We aim to complete human fashion outfit using a neural network that learns style compatibilities of its items. Our architecure is adapted from Learning Fashion Compatibility with Bidirectional LSTMs. and comprises 3 main components:

• A Cnn Encoder : Resnet18 pretrained on ImageNet
• A Single layer BiLSTM
• A MLP decoder with 2 hidden layers

• We perform training and evalutation of the poylvore dataset

By @waguy02