Model Versions

meme_model_v1.h5 - First meme model trained on subset of 20 images (~3800 captions). Had a softmax dense layer to filter the important features from the image/label embedding. Used 30 as maxlen. meme_model_lstm_attn.h5