I want to fine tune the good work in ImageNet-LT.
However, when I reloaded the provided checkpoint (ResNet50), I found the learning rate is 0.00000065.
But if I retrain from epoch 0, the learning rate should be 0.00395612 at epoch 400, according to the provided scripts (./sh/LT/ImageNetLT_train_R50.sh).
I want to fine tune the good work in ImageNet-LT. However, when I reloaded the provided checkpoint (ResNet50), I found the learning rate is 0.00000065. But if I retrain from epoch 0, the learning rate should be 0.00395612 at epoch 400, according to the provided scripts (./sh/LT/ImageNetLT_train_R50.sh).
Why is it the such value? Thanks!