Open WangYZ1608 opened 11 months ago
I am having similar results, any update?
Almost the same promblem. When I try to finetune the MAE model with MSE loss on a new small dataset, I found the loss was decreasing but the visualization results were quite strange (be like what you show above). So weired.
Hello,
thank you for the great work and the great repo.
I reproduced the pre-training of MAE-ViT-Large, performed 800 epochs, and set
norm_pix_loss=True
. Below is the visualization result and tensorboard curve, is this normal?the result similar to #156