def get_mean_and_std(dataset):
    '''Compute the mean and std value of dataset.'''
    dataloader = DataLoader(dataset, batch_size=1, shuffle=True, num_workers=2)
    mean = torch.zeros(3)
    std = torch.zeros(3)
    print('======> Computing mean and std..')
    for inputs, targets in dataloader:
        for i in range(3):
            mean[i] += inputs[:, i, :, :].float().mean()
            std[i] += inputs[:, i, :, :].float().std()
    mean.div_(len(dataset))
    std.div_(len(dataset))
    return mean, std