Is the text encoder in TMR fine-tuned using the BERT model, or is it retrained using a transformer?

[LinghaoChan]

It is a pre-trained model + transformer layers architecture. Please refer to the technical details in the appendix.

[caochengchen]

It is a pre-trained model + transformer layers architecture. Please refer to the technical details in the appendix.

I didn't see what text encoder is used in TMR in the appendix. I only saw that you mentioned sBERT in the paper. Is the text encoder in TMR fine-tuned using the BERT model, or is it retrained using a transformer?

[LinghaoChan]

@caochengchen I stated that it is extended from the text encoder of TEMOS. TEMOS use the DistilBERT + tf layers as the text encoder.

[caochengchen]

@caochengchen I stated that it is extended from the text encoder of TEMOS. TEMOS use the DistilBERT + tf layers as the text encoder.

Thank you very much