Closed kantenboki closed 2 years ago
image_snapshot_ticks
will use whatever value --snap
has (as seen here), so if you don't change it, it will use the default value of 50
. So, if you want to save both the model and the image snapshots more often, you must add --snap=10
.
Alternatively, you can decouple them and e.g. save the image snapshots more often than the network snapshots, as the former will indicate to you better how training is doing and you can save it as often as you want compared to the model snapshots. This is what I do in my StyleGAN3 repository, for example, so I can save the image snapshots every tick (--img-snap=1
), and save the model every 10 ticks (--snap=10
).
Thank you comments PDillis !! --snap=10 solved my problem.
If I wanted to set a number for each x and y 219 c.image_snapshot_ticks = c.network_snapshot_ticks = opts.snap → c.image_snapshot_ticks = 1 219+ c.network_snapshot_ticks = 50 is OK ???
Yes, but just keep in mind that you really shouldn't modify the values in the code itself, but as the parameters you input via the command line. As such, if you want to have two parameters, one for each different snapshot, then you should add the respective command in the code. That's what I did in my repo, so you could just use that one (or do it yourself in yours).
THANK YOU!!
I want to change image_snapshot_ticks=50 to 10.
So, I changed the image_snapshot_ticks in training_loop.py,
but when I look at trainig_options.json, the number have not changed.
How can I change the image_snapshot_ticks?