Crash while trying to upscale <16px per dimention images

[MikiP98]

Tested with 4x4 and 8x8 image.
After trying to upscale the 4x4 image by a factor of 2, the program crashed with:

ValueError: could not broadcast input array from shape (4,4,3) into shape (15,4,3)

on line:

new_img[0:pad_size, pad_size:-pad_size, :] = np.flip(image[0:pad_size, :, :], axis=0) #top

in file:

"{...}\site-packages\RealESRGAN\utils.py"

Image file:

Test code:

# coding=utf-8
import PIL.Image
import torch

from RealESRGAN import RealESRGAN


device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

image = PIL.Image.open("../../input/16-bit-P3-blue.png")

model = RealESRGAN(device, scale=2)
model.load_weights(f'weights/RealESRGAN_x{2}.pth')
image = model.predict(image)

As a temporal workaround I'm uscaling all input images to the minimum size of 16px for the smaller dimention using PIL scaling.
I haven't tested this much, so IDK if this workaround is fullproof. It may require minumum size of 8 or 32 for a scaling factor of 4 as far as I know.
I will return to this if I ever test it more.