Ming-Hsuan-Tu

this is very helpful. thank you. Another question is that how you select the learning rate scheduler? I tried to train resnet18 using AdamP on CIFAR100 with step decay scheduler,...

@TalHassner Can you elaborate more detail?

@TalHassner as suggested in https://github.com/dougsouza/face-frontalization/issues/16, I tried to project 3D landmark using the same projection matrix. ``` threedee2 = np.reshape(lmk, (-1, 3), order='F').transpose() temp_proj2 = proj_matrix * np.vstack((threedee2, np.ones((1, threedee2.shape[1]))))...

@TalHassner yup. I have read the paper last week, but still can't figure out how to do the mapping without the index of ref_U. by the way, I found that...

@zongoalbert Where did you get this?

what did you mean `loss scale`? Is it related to classification loss and regression loss?

so you mean that if the classification loss is 0.3, then smallest distillation loss (one of the fpn level) should approach to 0.3 as well?

I found it may be sensitive to these parameters in my own dataset, since there is no improvement and even worse than the baseline. I would follow your suggestion and...

@yzd-v Do you mind to upload your training log?

Here are my distiller loss Are these losses too small? the loss value for different fpn levels differ a lot.