Initialize DecoderCNN in Image captioning #218
Description
The performance of the model with those default hyper parameters take larger time to train and unusually larger loss at starting point. Can you please tell how to initialize the DecoderCNN using Xavier initialization?