Can you provide the detail in finetuning the model?
Opened this issue · 0 comments
ZhuYi98 commented
I downloaded the ResNet18 ckpt and used the LinearClassifier head on top of the Resnet model as provided in eval_moco_ins.py, the learning rate is 0.01 and 0.0002 for LinearClassifier and Resnet, the optimizer is SGD as default. I only achieved an average AUROC of around 75 for ChestX-ray14 finetuning instead of 83.5 in the paper.
I wonder what went wrong and want to know your fintuning details. Thank you.