BERT_multimodal_transformer issues

BERT is better than BERT+MAG

I simply return text_embedding in MAG's forward() function to reproduce performance of BERT model `def forward(self, text_embedding, visual, acoustic): return text_embedding` but I found that on MOSI, BERT model 5...

Norwa9

About results on MOSEI

1

Is acc_2 result on MOSEI dataset calculated without label=0 examples ?

yingtaoHuo

about seven-class accuracy (Acc- 7)

Can you tell me why the Acc-7 is not reported in your paper?

fly-pandas

AttributeError: 'MAG_BertModel' object has no attribute 'get_extended_attention_mask'

Dear Rahman, when I reproduce the [code](https://github.com/WasifurRahman/BERT_multimodal_transformer/blob/dc7876fc30f7ef362999200911e3d4d8a2bca107/bert.py#L180) it seems that there has an error when use the get_extended_attention_mask , can you give me some advice?

BaiMeiyingxue