To continue my train.

[john-rocky]

I want to let my model read my checkpoint and continue my train. How can I let my model read the checkpoint ?

[benckx]

Normally if the checkpoint is in the folder 'checkpoint', it will be picked up automatically when you resume training. But you need to use the same parameters (batch_size, image size, etc.)

It will create the first checkpoint after 500 iterations or something.

[hi0001234d]

@john-rocky normally it works like this and @benckx is correct but i've had one confusion recently where epoch starts from beginning even after loading checkpoints which i think is not normal however i must admint that i've_ not long experience in reusing checkpoints.

[fooness]

@john-rocky normally it works like this and @benckx is correct but i've had one confusion recently where epoch starts from beginning even after loading checkpoints which i think is not normal however i must admint that i've_ not long experience in reusing checkpoints.

Could anyone please verify if the described behavior is how “continue training” should work?

My stdout looks e.g. like this:

 [*] Reading checkpoints...
 [*] Success to read DCGAN.model-2
 [*] Load SUCCESS
Epoch: [ 0/ 1] [   0/ 220] time: 14.7326, d_loss: 623.09069824, g_loss: 0.00000000

So it starts countint again from 0 …

[madmannnn]

@john-rocky normally it works like this and @benckx is correct but i've had one confusion recently where epoch starts from beginning even after loading checkpoints which i think is not normal however i must admint that i've_ not long experience in reusing checkpoints.

Could anyone please verify if the described behavior is how “continue training” should work?

My stdout looks e.g. like this:

 [*] Reading checkpoints...
 [*] Success to read DCGAN.model-2
 [*] Load SUCCESS
Epoch: [ 0/ 1] [   0/ 220] time: 14.7326, d_loss: 623.09069824, g_loss: 0.00000000

So it starts countint again from 0 …

I met the same problem. Is it normal?