kxngluca/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Pull Requests Activity

Allow attention to be specified for stylegan2

Browse Source
This commit is contained in:
James Betker 2020-11-14 09:29:53 -07:00
parent cdc5ac30e9
commit 423ee7cb90
2 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
codes/models/archs/stylegan2.py
View File

@ -537,7 +537,7 @@ class Generator(nn.Module):
for style, block, attn in zip(styles, self.blocks, self.attns): for style, block, attn in zip(styles, self.blocks, self.attns):
if exists(attn): if exists(attn):
x = attn(x) x = checkpoint(attn, x)
x, rgb = checkpoint(block, x, rgb, style, input_noise, structure_input) x, rgb = checkpoint(block, x, rgb, style, input_noise, structure_input)
return rgb return rgb

7
codes/models/networks.py
View File

@ -134,8 +134,10 @@ def define_G(opt, net_key='network_G', scale=None):
netG = LinearLatentEstimator(in_nc=3, nf=opt_net['nf']) netG = LinearLatentEstimator(in_nc=3, nf=opt_net['nf'])
elif which_model == 'stylegan2': elif which_model == 'stylegan2':
is_structured = opt_net['structured'] if 'structured' in opt_net.keys() else False is_structured = opt_net['structured'] if 'structured' in opt_net.keys() else False
attn = opt_net['attn_layers'] if 'attn_layers' in opt_net.keys() else []
netG = StyleGan2GeneratorWithLatent(image_size=opt_net['image_size'], latent_dim=opt_net['latent_dim'], netG = StyleGan2GeneratorWithLatent(image_size=opt_net['image_size'], latent_dim=opt_net['latent_dim'],
style_depth=opt_net['style_depth'], structure_input=is_structured) style_depth=opt_net['style_depth'], structure_input=is_structured,
attn_layers=attn)
else: else:
raise NotImplementedError('Generator model [{:s}] not recognized'.format(which_model)) raise NotImplementedError('Generator model [{:s}] not recognized'.format(which_model))
return netG return netG
@ -195,7 +197,8 @@ def define_D_net(opt_net, img_sz=None, wrap=False):
elif which_model == "pyramid_disc": elif which_model == "pyramid_disc":
netD = SRGAN_arch.PyramidDiscriminator(in_nc=3, nf=opt_net['nf']) netD = SRGAN_arch.PyramidDiscriminator(in_nc=3, nf=opt_net['nf'])
elif which_model == "stylegan2_discriminator": elif which_model == "stylegan2_discriminator":
disc = StyleGan2Discriminator(image_size=opt_net['image_size'], input_filters=opt_net['in_nc']) attn = opt_net['attn_layers'] if 'attn_layers' in opt_net.keys() else []
disc = StyleGan2Discriminator(image_size=opt_net['image_size'], input_filters=opt_net['in_nc'], attn_layers=attn)
netD = StyleGan2Augmentor(disc, opt_net['image_size'], types=opt_net['augmentation_types'], prob=opt_net['augmentation_probability']) netD = StyleGan2Augmentor(disc, opt_net['image_size'], types=opt_net['augmentation_types'], prob=opt_net['augmentation_probability'])
else: else:
raise NotImplementedError('Discriminator model [{:s}] not recognized'.format(which_model)) raise NotImplementedError('Discriminator model [{:s}] not recognized'.format(which_model))