Python Networks Examples

Example #1

File: train_E_1024_kl.py Project: disanda/PGGAN-TA

# netG = Networks.Generator(depth=9, latent_size=512)
# z = z.squeeze(2).squeeze(2)
# x_ = netG(z,depth=8,alpha=1)
# print(z.shape)
# print(x_.shape)

#----------------test pre-model output-----------


def toggle_grad(model, requires_grad):
    for p in model.parameters():
        p.requires_grad_(requires_grad)


netG = torch.nn.DataParallel(
    net.Generator(depth=9, latent_size=512)
)  # in: [-1,512], depth:0-4,1-8,2-16,3-32,4-64,5-128,6-256,7-512,8-1024
netG.load_state_dict(
    torch.load('./pre-model/GAN_GEN_SHADOW_8.pth',
               map_location=device))  #shadow的效果要好一些
netD1 = torch.nn.DataParallel(
    net.Discriminator(height=9, feature_size=512)
)  # in: [-1,3,1024,1024],out:[], depth:0-4,1-8,2-16,3-32,4-64,5-128,6-256,7-512,8-1024
netD1.load_state_dict(
    torch.load('./pre-model/GAN_DIS_8.pth', map_location=device))

netD2 = torch.nn.DataParallel(Encoder.encoder_v1(height=9, feature_size=512))
#netD2 = torch.nn.DataParallel(Encoder.encoder_v2()) #新结构，不需要参数
toggle_grad(netD1, False)
toggle_grad(netD2, False)

Example #2

File: train_E4G.py Project: disanda/TA-V2

    os.mkdir(resultPath1_1)

resultPath1_2 = resultPath + "/models"
if not os.path.exists(resultPath1_2):
    os.mkdir(resultPath1_2)

#----------------test pre-model output-----------


def toggle_grad(model, requires_grad):
    for p in model.parameters():
        p.requires_grad_(requires_grad)


netG = torch.nn.DataParallel(
    net.Generator(depth=9, latent_size=512)
)  # in: [-1,512], depth:0-4,1-8,2-16,3-32,4-64,5-128,6-256,7-512,8-1024
netG.load_state_dict(
    torch.load('./pre-model/GAN_GEN_SHADOW_8.pth',
               map_location=device))  #shadow的效果要好一些

netE = torch.nn.DataParallel(Encoder.encoder_v1(height=9, feature_size=512))
#netE.load_state_dict(torch.load('./pre-model/D2E_std_L2_ep9.pth',map_location=device))
netE.load_state_dict(
    torch.load('./pre-model/D_all_Loss_ep19.pth', map_location=device))

#-------------load single image--------------
loader = torchvision.transforms.Compose([torchvision.transforms.ToTensor()])

from PIL import Image

Example #3

# print(type(dict2))

# keys = []
# dict3 = {}

# for i,j in dict1.items():
# 	if i.startswith('fc'):
# 		print(i)
# 		continue
# 	keys.append(i)

# dict3 = {k:dict1[k] for k in keys}
# print(dict3.keys())

from pro_gan_pytorch import Encoder, Networks as net
netG1 = torch.nn.DataParallel(net.Generator(depth=9, latent_size=512)).to(
    device
)  # in: [-1,512], depth:0-4,1-8,2-16,3-32,4-64,5-128,6-256,7-512,8-1024
#netG1.load_state_dict(torch.load('D:\\AI-Lab\\PGGAN-TA\\result\\Step2_Training_EL2_GL2\\models\\G_model_ep0.pth',map_location=device))
#netG1.load_state_dict(torch.load('./pre-model/GAN_GEN_SHADOW_8.pth',map_location=device))
#netG1.load_state_dict(torch.load('D:\\AI-Lab\\PGGAN-TA\\result\\Step2_G-allLoss-allLoss_wwm\\models\\G_model_ep3.pth',map_location=device))
netG1.load_state_dict(
    torch.load(
        'D:\\AI-Lab\\PGGAN-TA\\result\\Step2_G_wwm_allLoss\\models\\G_model_ep8.pth',
        map_location=device))
netEn = torch.nn.DataParallel(Encoder.encoder_v1(height=9, feature_size=512))
#netEn.load_state_dict(torch.load('./pre-model/D2E_std_L2_ep9.pth',map_location=device))
netEn.load_state_dict(
    torch.load('./pre-model/D_all_Loss_ep19.pth', map_location=device))

#--------------操作 y-> z -> x-------------

Example #4

resultPath1_2 = resultPath + "/models"
if not os.path.exists(resultPath1_2):
    os.mkdir(resultPath1_2)

#----------------pre-model-----------


def toggle_grad(model, requires_grad):
    for p in model.parameters():
        p.requires_grad_(requires_grad)


in_dim = 512

netG1 = torch.nn.DataParallel(
    net1.Generator(depth=9, latent_size=in_dim)
)  # in: [-1,512], depth:0-4,1-8,2-16,3-32,4-64,5-128,6-256,7-512,8-1024
netG1.load_state_dict(
    torch.load('./pre-model/GAN_GEN_SHADOW_8.pth',
               map_location=device))  #shadow的效果要好一些
netD1 = torch.nn.DataParallel(
    net1.Discriminator(height=9, feature_size=in_dim)
)  # in: [-1,3,1024,1024],out:[], depth:0-4,1-8,2-16,3-32,4-64,5-128,6-256,7-512,8-1024
netD1.load_state_dict(
    torch.load('./pre-model/GAN_DIS_8.pth', map_location=device))

netG2 = torch.nn.DataParallel(net2.Decoder_v1(depth=9, latent_size=in_dim))
netD2 = torch.nn.DataParallel(net2.Encoder_v1(height=9,
                                              feature_size=in_dim))  #新结构，不需要参数

toggle_grad(netD1, False)