Dear scholar,
Because of the LXMERT pretrains on GQA train +validation datasets, it is not consistent of your settings where you remove the validation split from your pretraining datasets. So do you provide the pretrained model parameters from the updated datasets( removing the validation datasets)？ And what is your hypermeter for LXMERT. I want to reproduce your paper's results.
Dear scholar, Because of the LXMERT pretrains on GQA train +validation datasets, it is not consistent of your settings where you remove the validation split from your pretraining datasets. So do you provide the pretrained model parameters from the updated datasets( removing the validation datasets)？ And what is your hypermeter for LXMERT. I want to reproduce your paper's results.