utils.py

import numpy as np
from torchvision import utils


def make_numpy_grid(tensor_data, pad_value=0, padding=0):
    tensor_data = tensor_data.detach()
    vis = utils.make_grid(tensor_data, pad_value=pad_value, padding=padding)
    vis = np.array(vis.cpu()).transpose((1, 2, 0))
    if vis.shape[2] == 1:
        vis = np.stack([vis, vis, vis], axis=-1)
    return vis


def de_norm(tensor_data):
    return tensor_data * 0.5 + 0.5