Custom training a StyleGAN2 model

In this video you’re learn the steps needed to train your own StyleGAN2 model. I walk you thru uploading your image dataset, converting it to .tfrecords files, and finally running the training command.

Three other videos worth watching:

[galabyca]

6:16 damn you scared me! good tutorial, thx!

[dumbcruel]

wow so glad im getting this in my suggested videos

[chitti]

how do i fix: line 291, in upsample_conv_2d
x = tf.nn.conv2d_transpose(x, w, output_shape=output_shape, strides=stride, padding='VALID', data_format=data_format),
TypeError: int() argument must be a string, a bytes-like object or a number, not 'Tensor'

[justinhuang]

Serious gem man. I"m working on a capstone about this. This is going to help so much.

[dwaejishan]

i just did a scraping from instagram and those image are sized 1080x1080, can i use 1080x1080 images instead of 1024x1024?

[jeremykeenan]

I get the following error when training - any ideas?

NotImplementedError: Cannot convert a symbolic Tensor (Inputs/minibatch_gpu_in:0) to a numpy array.

[sorenaide]

nice video but maybe try to fix the earrape at 6:16

[jochemstoel]

Hey, this is great thanks. Could you make a Jupyter notebook for this?

[ahmedshingaly]

very good tutorial, thank you
i am having this problem in the beginning of training

fail at cwise_ops_common and
from device: CUDA_ERROR_OUT_OF_MEMORY: out of memory
any idea on how to prevent memory leak?
I reduced minibatch size but still getting same problem

[harshjain]

Where do we get the zipped data files from?

[jonathynblythe]

Is there any way StyleGAN can be downloaded as a software application for one's own desktop? I've seen videos of github and all that along with the python language and downloading other applications. But for the layman, none of this makes any sense. We're used to downloading software with .exe in it. The tutorials I've looked at for github stuff lose me at about half way through. Thanks.

[lfdnjpy]

Hey, im trying to follow your workflow and im very excited to try it out by myself. But i can't find the video for the next step? What do i have to do with the .pkl file to generate images from my trained model?

[DigitalMarksman]

Hey, great video! Thank you for your channel and the job you do here :)
Can I ask some sort of question, in your videos you say that there is a way to use 2 GPUs, like P100 or V100, so the question is what would be the most optimal settings for running on them ?
I am particularly interested in "minibatch_size_base" and "minibatch_gpu_base", which by default have values of 32 and 4 respectively. Can you please suggest anything about it ?

[gayrattangriberganov]

Hi bro
lets say if your dataset contains 256px img, then StyleGAN anyway generates 1024 px img after finishing train?

[jordan]

Would you mind what version of CUDA and cdnn are you running during this tutorial ?

[JoshBrownKramer]

When I train, it starts from scratch. How do I transfer learn, as it appears you are doing?

[MohamedHassan-sowy]

Awesome tutorial :D
When I tried to run it with another square-shape 512x512 dogs dataset and run it on 8 RTX 2080 Ti GPUs, after 12, 332 Kimg length and around 300 ticks, I tried to test one checkpoint to see if it would work or not. But it didn't work it always generates black photos with different seeds values.
Could you help with that?
Note that I've used a dataset from Kaggle and I used dataset_tool.py to prepare it and I managed to display it.

[harveymogarawanderingfilip]

How many RAMs needed to do this? I want to try it

[berkozdemir]

Hi bustbright, thank you for these great videos!


I made my own dataset with 32x32 images, and I get this error when I want to train them:


Dataset shape = [3, 32, 32]
Dynamic range = [0, 255]
Label size = 0
[]
Traceback (most recent call last):
File "run_training.py", line 198, in <module>
main()
File "run_training.py", line 193, in main
run(**vars(args))
File "run_training.py", line 126, in run
dnnlib.submit_run(**kwargs)
File "C:\Users\berko\Desktop\stylegan2\stylegan2\dnnlib\submission\submit.py", line 343, in submit_run
return farm.submit(submit_config, host_run_dir)
File "C:\Users\berko\Desktop\stylegan2\stylegan2\dnnlib\submission\internal\local.py", line 22, in submit
return run_wrapper(submit_config)
File "C:\Users\berko\Desktop\stylegan2\stylegan2\dnnlib\submission\submit.py", line 280, in run_wrapper

File "C:\Users\berko\Desktop\stylegan2\stylegan2\training\training_loop.py", line 149, in training_loop
resume_pkl, resume_kimg =
File "C:\Users\berko\Desktop\stylegan2\stylegan2\training\misc.py", line 56, in locate_latest_pkl
latest_pickle = allpickles[-1]
IndexError: list index out of range


googling didn't help, apparently I am not very experienced with these stuff :(

[Tokyyto]

Thank you!! What would be the difference of training it with Runway?