loss and acc curves
andriyrizhiy opened this issue · 7 comments
Can i get your loss and acc curves during training?
Sorry, I did not save the loss and acc curves in the CVPR version. In that version, I almost did not use data augmentation, so the training loss is much lower than the validation loss.
In the extended journal version, I do have the loss curves as follows:
The base model is also changed in the journal version. But the loss cuvres have the same tendancy.
in your experiment val acc also stood still like val loss curve?
becouse i get like this
for acc4/5 (train is green and val is gray)
It is not used ROD (only roi_align)
I did not plot the acc curves. But they should have the similar tendency as the loss curves.
what is end value of you train acc? it is much more than val acc?
Yes, the training acc is higher than the val acc.
but not too mush