it seems this code doesn‘t generate M sentences for each image during training (Eq. 3 in the paper)

[winnechan]

hi, as a fresher in research, i have read your paper recently and have been wondering how to generate M different sentences for one given image without involving random noise or non-argmax sampling in the model proposed in your paper. also, it seems this code doesn‘t generate M sentences for each image during training.

pls help
thx