Junction4Nako/mvp_pytorch

Question about id2phrase.json

Opened this issue · 1 comments

Hello, could you please tell me how the format of id2phrase.json file is generated?
What is the meaning of the data like "30522": [" Boat ", "two"] in that file?
Is the data in it stored as "image_id: tuple"?
Or does "30522" represent a randomly generated num?

30522 represents the embedding id of the phrase ["boat", "two"], which corresponds to the 30522nd embedding of the bert.embedding.word_embedding layer.