Training StyleGAN2-ADA Class: Week 2 (ADA changes, Training Parameters, What to Watch Out For)

[@isoscelesVJ]

Amazing explanation! This answers so many questions that I have been continually running into. Many thanks for sharing publicly.

[@ArtificialImages]

glad it helped!

[@apachidesigns]

17:40 - "Right on cue. The cat". Brilliant :-)

[@omarguzman1112]

Can you please post the colab notebook?

[@saisandeepkantareddy1890]

Hi, thanks for the video, what's the snapshot number, for example, snapshot 2419.pkl, does it says something related to kimg? I mean after 2419 kimg the fidl 50k full score is 50? can I interpret like that?

[@Cauthon007]

Hi out of curiosity what would be a recmmended cfg for a 512 image on a 24gb gpu (rtx3090) im just running on auto right now as i daren't change it without speaking to someone like yourself

[@nithyakannan7971]

i want to find fid score but it runs continuously still doesn't find any score. What can be done? Also is there any other way to measure the performance?

[@ricardocesarrodrigues837]

in "Train.py --help" you may find how to take metrics during training, or just train it with --metrics None, than use calc_metrics.py to get the score you want.

[@troonics]

I used colab for training and after i got disconnected, I upload my .pkl file from my drive it stops after one tick and says ^C. What dont it start training again?

[@ArtificialImages]

what GPU are you being assigned? its possible you’re maxing out your VRAM

[@troonics]

@@ArtificialImages yes sir, I was assigned Tesla k80 and ram was 12 gb

[@troonics]

I worked for 4 tick in 4 hours and when I reconnected with second picked pkl file it just stops after 1 tick

[@ArtificialImages]

@@troonics yes that might be the issue then. I’d try to get a p100 or v100 in order to train on colab

[@Movieboy777]

@@ArtificialImages i had the same issue and it persisted even when i got a p100 .. : / .. is there maybe a way of limiting the amount of VRAM the process uses so that it doesn't max out ?