- 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