google / seq2seq

A general-purpose encoder-decoder framework for Tensorflow

Home Page:https://google.github.io/seq2seq/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

target_ids requirement for image2seq

changt opened this issue · comments

If I don't input image/caption_ids for my tfRecord in the ImageCaptioningInputPipeline, the model won't train and will complain. However, it looks like the preprocessor is setting the target_ids field inside Image2Seq. What exactly is image/caption_ids and why is it required?