I'm curious about the performance of the textual stream.

[daewon88]

Dear @nicolalandro

Hello! Thank you for sharing your great work.

I read your paper, and I tried to reproduce the result of textual stream using Bert model as suggested in the paper. However, in my case the performance of classification is about 25%.

Do you know what the issue is? Or do you have any tips on textual model training? Thanks for the help in advance! Kind regards, Daewon