Closed JackieZhangdx closed 7 years ago
Hi, that's normal. The parameter randomization in the augmentation layer can lead to invalid augmentation proposals. The layer will try a few times to find a valid augmentation, but there's a cutoff because searching for too long would slow down training.
Truly grateful for your help.
Hi.
I try to train flownet on Sintel dataset and modify the crop size parameter in augmentation layer. In my training process, there are several warning :" augmentation_layer_base.cpp:166] Augmentation: Exceeded maximum tries in finding spatial coeffs." The warning comes 3~10/100 iter.
I'm wondering if the warning would influence the training results.
Tks a lot!