search

DeepRNN / image_captioning

Tensorflow implementation of "Show, Attend and Tell: Neural Image Caption Generation with Visual Attention"
MIT License
781 stars 351 forks source link

The evaluation results are very strange #52

Open Kinghup opened 5 years ago

Kinghup commented 5 years ago

I use the trained model to evaluate the score and the generated sentence is reasonable. But the score of BLEU is very strange. If anyone have ideas? image

chansongoal commented 5 years ago

Have you solve the problem? I have also enchanted the same problem with you. Are you willing to talk about it with me? Looking forward to your reply. Thank you. You can contact me through my email:gcs@mail.ustc.edu.cn