54 lines
3.1 KiB
Plaintext
54 lines
3.1 KiB
Plaintext
Usage: train.py [OPTIONS]
|
|
|
|
Train a GAN using the techniques described in the paper "Alias-Free
|
|
Generative Adversarial Networks".
|
|
|
|
Examples:
|
|
|
|
# Train StyleGAN3-T for AFHQv2 using 8 GPUs.
|
|
python train.py --outdir=~/training-runs --cfg=stylegan3-t --data=~/datasets/afhqv2-512x512.zip \
|
|
--gpus=8 --batch=32 --gamma=8.2 --mirror=1
|
|
|
|
# Fine-tune StyleGAN3-R for MetFaces-U using 1 GPU, starting from the pre-trained FFHQ-U pickle.
|
|
python train.py --outdir=~/training-runs --cfg=stylegan3-r --data=~/datasets/metfacesu-1024x1024.zip \
|
|
--gpus=8 --batch=32 --gamma=6.6 --mirror=1 --kimg=5000 --snap=5 \
|
|
--resume=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-ffhqu-1024x1024.pkl
|
|
|
|
# Train StyleGAN2 for FFHQ at 1024x1024 resolution using 8 GPUs.
|
|
python train.py --outdir=~/training-runs --cfg=stylegan2 --data=~/datasets/ffhq-1024x1024.zip \
|
|
--gpus=8 --batch=32 --gamma=10 --mirror=1 --aug=noaug
|
|
|
|
Options:
|
|
--outdir DIR Where to save the results [required]
|
|
--cfg [stylegan3-t|stylegan3-r|stylegan2]
|
|
Base configuration [required]
|
|
--data [ZIP|DIR] Training data [required]
|
|
--gpus INT Number of GPUs to use [required]
|
|
--batch INT Total batch size [required]
|
|
--gamma FLOAT R1 regularization weight [required]
|
|
--cond BOOL Train conditional model [default: False]
|
|
--mirror BOOL Enable dataset x-flips [default: False]
|
|
--aug [noaug|ada|fixed] Augmentation mode [default: ada]
|
|
--resume [PATH|URL] Resume from given network pickle
|
|
--freezed INT Freeze first layers of D [default: 0]
|
|
--p FLOAT Probability for --aug=fixed [default: 0.2]
|
|
--target FLOAT Target value for --aug=ada [default: 0.6]
|
|
--batch-gpu INT Limit batch size per GPU
|
|
--cbase INT Capacity multiplier [default: 32768]
|
|
--cmax INT Max. feature maps [default: 512]
|
|
--glr FLOAT G learning rate [default: varies]
|
|
--dlr FLOAT D learning rate [default: 0.002]
|
|
--map-depth INT Mapping network depth [default: varies]
|
|
--mbstd-group INT Minibatch std group size [default: 4]
|
|
--desc STR String to include in result dir name
|
|
--metrics [NAME|A,B,C|none] Quality metrics [default: fid50k_full]
|
|
--kimg KIMG Total training duration [default: 25000]
|
|
--tick KIMG How often to print progress [default: 4]
|
|
--snap TICKS How often to save snapshots [default: 50]
|
|
--seed INT Random seed [default: 0]
|
|
--fp32 BOOL Disable mixed-precision [default: False]
|
|
--nobench BOOL Disable cuDNN benchmarking [default: False]
|
|
--workers INT DataLoader worker processes [default: 3]
|
|
-n, --dry-run Print training options and exit
|
|
--help Show this message and exit.
|