Hi,
I have two questions:
First : Why didn't you use K Fold Cross Validation?
Second : What is the reason use different learning rate for classifier? Is it for faster convergence?
I am trying to adapt CSRA to EfficientNetB3 on my multi-label dataset. Although I try various head and lambda numbers, I am getting worse results according to baseline model. What is your opinion? Is there also something different to try?
Also there is class imbalance in my dataset. Is there need to make data augmentation to prevent class imbalance? Is CSRA a method affected by data augmentation?
Thanks
Hi,
I have two questions:
First : Why didn't you use K Fold Cross Validation?
Second : What is the reason use different learning rate for classifier? Is it for faster convergence?
I am trying to adapt CSRA to EfficientNetB3 on my multi-label dataset. Although I try various head and lambda numbers, I am getting worse results according to baseline model. What is your opinion? Is there also something different to try?
Also there is class imbalance in my dataset. Is there need to make data augmentation to prevent class imbalance? Is CSRA a method affected by data augmentation?
Thanks