utils.py

import os
import torch
from torch.autograd import Variable
from torch.nn import init

def make_folder(path, version):
        if not os.path.exists(os.path.join(path, version)):
            os.makedirs(os.path.join(path, version))


def tensor2var(x, grad=False):
    if torch.cuda.is_available():
        x = x.cuda()
    return Variable(x, requires_grad=grad)

def var2tensor(x):
    return x.data.cpu()

def var2numpy(x):
    return x.data.cpu().numpy()

def denorm(x):
    out = (x + 1) / 2
    return out.clamp_(0, 1)

def weights_init(m):
    classname = m.__class__.__name__
    # print(classname)
    if classname.find('Conv2d') != -1:
        init.xavier_normal_(m.weight.data)
        init.constant_(m.bias.data, 0.0)
    elif classname.find('Linear') != -1:
        init.xavier_normal_(m.weight.data)
        init.constant_(m.bias.data, 0.0)