zhjohnchan / M3AE

[MICCAI-2022] This is the official implementation of Multi-Modal Masked Autoencoders for Medical Vision-and-Language Pre-Training.
115 stars 11 forks source link

Split for VQARAD #3

Closed GanjinZero closed 2 years ago

GanjinZero commented 2 years ago

It seems that others use VQARAD with a train (3064) + test (451) set. And it seems you have a dev set for VQARAD? What is your split config?

GanjinZero commented 2 years ago

Also why C4_abnormal is not used in MedVQA2019 train / dev / test?

zhjohnchan commented 2 years ago

Hi, following the settings of previous studies, I used the commonly used split (3064 + 451). I just duplicated the test set to unify the code interface. For MedVQA-2019, I followed this paper (https://www.nature.com/articles/s41598-021-98390-1) to set up the experiments and did not include C4_abnormal.

GanjinZero commented 2 years ago

got it