search

SydCaption / SAAT

MIT License
62 stars 21 forks source link

svo is correct but the caption is wrong #24

Closed ydwl-lynn closed 3 years ago

ydwl-lynn commented 3 years ago

Hello,I reproduced the results of your paper, and I found that some of the video output svo is correct but the caption is wrong. What is the reason?If it is convenient, I would like to discuss further with you.Thank you very much.

SydCaption commented 3 years ago

Hi, I'd welcome any discussion. You're right, svo outcome can only partially adjust the decoder instead of controlling it completely. I think there are two reasons, 1) decoder still heavily relies on linguistic prior, e.g., word co-occurrence, to optimize the gram-level cross-entropy loss; 2) svo prediction is still far from satisfactory.

ydwl-lynn commented 3 years ago

Thank you very much for your reply。I still have some details that I would like to discuss with you. If it is convenient, please send your WeChat to my mailbox（1007369109@qq.com）。Thank you very much!