def generate():
opt = Config()
criterion = nn.BCELoss()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")# 训练可能多卡，预测一张就够了，所以有点小不同
#dnet = torch.load('dnet1.pth').to(device)#可能需要从其他GPU移动到0号，若满足条件则不作为
#gnet = torch.load('gnet1.pth').to(device)
dnet = DNet(opt).to(device)
gnet = GNet(opt).to(device)
state_dict = torch.load('dd.pth')
new_state_dict = OrderedDict()
for k,v in state_dict.items():
name = k[7:]
new_state_dict[name] = v
dnet.load_state_dict(new_state_dict)
state_dict = torch.load('gg.pth')
new_state_dict = OrderedDict()
for k,v in state_dict.items():
name = k[7:]
new_state_dict[name] = v
gnet.load_state_dict(new_state_dict)
dnet.eval()
gnet.eval()
noise = torch.randn(opt.batch_size, opt.nd, 1, 1, device=device)
#with torch.no_grad():
fake = gnet(noise)
output = dnet(fake)
label = torch.full((opt.batch_size, ), opt.real_label, device=device)
d_err_fake = criterion(output, label) # 生成图像的损失；还是tensor
mean_score = output.mean() #生成图像的平均得分；还是tensor
fake_img = vutils.make_grid(fake, normalize=True)
writer = SummaryWriter(log_dir='generate_rusult')
writer.add_image('fake_img', fake_img)
writer.close()
print('生成图像的平均损失值：%.4f'%d_err_fake.item())
print('生成图像的平均得分：%.4f'%mean_score.item())
def generate(opt, device):
criterion = nn.BCELoss()
dnet = DNet(opt).to(device) # 可能需要从其他GPU移动到0号，若满足条件则不作为
gnet = GNet(opt).to(device)
state_dict = torch.load('dnet.pth')
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[7:] # remove 'module.'
new_state_dict[name] = v
dnet.load_state_dict(new_state_dict)
state_dict = torch.load('gnet.pth')
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[7:]
new_state_dict[name] = v
gnet.load_state_dict(new_state_dict)
dnet.eval()
gnet.eval()
noise = torch.randn(opt.batch_size, opt.nd, 1, 1, device=device)
with torch.no_grad():
fake = gnet(noise)
output = dnet(fake)
label = torch.full((opt.batch_size, ), opt.real_label, device=device)
d_err_fake = criterion(output, label) # 生成图像的损失；还是tensor
mean_score = output.mean() #生成图像的平均得分；还是tensor
fake_img = vutils.make_grid(fake, normalize=True)
writer = SummaryWriter(log_dir='generate_result')
writer.add_image('fake_img', fake_img)
writer.close()
print('生成图像的平均损失值：%.4f' % d_err_fake.item())
print('生成图像的平均得分：%.4f' % mean_score.item())
metavar='M',
help='model file to load for evaluating.')
args = parser.parse_args()
# Model
model_gcn = GNet()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
state_dict = torch.load(args.load, map_location=device)
model_gcn.load_state_dict(state_dict)
# Turn batch norm into eval mode
# for child in model_gcn.feat_extr.children():
# for ii in range(len(child)):
# if type(child[ii]) == torch.nn.BatchNorm2d:
# child[ii].track_running_stats = False
model_gcn.eval()
# Cuda
use_cuda = torch.cuda.is_available()
if use_cuda:
model_gcn.cuda()
print('Using GPU')
else:
print('Using CPU')
# Graph
graph = Graph("./ellipsoid/init_info.pickle")
# Data Loader
folder = CustomDatasetFolder(args.data, extensions=["dat"], print_ref=False)
val_loader = torch.utils.data.DataLoader(folder, batch_size=1, shuffle=True)