Closed bareblackfoot closed 4 years ago
Hi, our word embedding is trained from scratch. The vocab.json is just a dict, which contains all words in Flickr30k. You can generate it by yourself.
Is it a JSON file using "node.idx"? Does it contain phrases like node.idx file or just a single word? for example. node.idx has "straddle bike". Maybe I can generate the file but the generated file can have its variation and affect the performance as i think...
Okay, I have uploaded this file. You can access it in Google Drive.
It's not easy for me to access Google! (you maybe do not understand this)
I think it is a directory for the word embeddings, but I couldn't find it from your uploaded dataset. Could you upload the skip-thoughts directory on your google drive?