search

CQFIO / PhotographicImageSynthesis

Photographic Image Synthesis with Cascaded Refinement Networks
https://cqf.io/ImageSynthesis/
1.25k stars 227 forks source link

weights lambda #10

Closed DirtyHarryLYL closed 7 years ago

DirtyHarryLYL commented 7 years ago

Thx for the great project first~ My question: In the train demo,what does " the weights lambda are collected at 100th epoch" mean? Should I change lambda in "tf.image.resize_area(label,(sp//16,sp//8)))*10/1.5" with new value got in training by hand? Sincerely

CQFIO commented 7 years ago

For the first question, the weights are collected by taking the inverse of the average of l0,..,l5 in the 100th epoch. For the second question, in demo_256p, I use the diversity loss and I use the simple loss in demo_512p and demo_1024p. I generate 9 output images in demo_256p.

DirtyHarryLYL commented 7 years ago

Thanks a lot~ p_i=compute_error(~~)*mean(l_i)_100th_epoch, i=0,1,...5