How do I create a scale matrix for rescaling a PyTorch tensor, and then how do I use it?

Question

I need to create a scale matrix that is autograd compatible, works on B,C,H,W tensors, and takes input values (possibly generated randomly) for controlling the scaling. How can I generate and use a scale matrix for this?

import torch
import torch.nn.functional as F
import torchvision.transforms as transforms
from PIL import Image


# Load image
def preprocess_simple(image_name, image_size):
    Loader = transforms.Compose([transforms.Resize(image_size), transforms.ToTensor()])
    image = Image.open(image_name).convert('RGB')
    return Loader(image).unsqueeze(0)
    
# Save image   
def deprocess_simple(output_tensor, output_name):
    output_tensor.clamp_(0, 1)
    Image2PIL = transforms.ToPILImage()
    image = Image2PIL(output_tensor.squeeze(0))
    image.save(output_name)


def get_scale_mat(theta):
    ...
    return scale_mat


def scale_img(x, theta, dtype):
    scale_mat = get_scale_mat(theta)

    # Can F.affine_grid & F.grid_sample be used with a scale matrix?
    grid = F.affine_grid(scale_mat , x.size()).type(dtype)
    x = F.grid_sample(x, grid)
    return x


# Shear tensor
test_input = # Test image
scale = 5 # Example value
scaled_tensor = scale_img(test_input, scale)

Answer 1

This is how you create and use a 3x2 scale matrix with F.affine_grid and F.grid_sample:

def get_scale_mat(m, device, dtype):
    scale_mat = torch.tensor([[m, 0., 0.],
                              [0., m, 0.]])
    return scale_mat
    
def scale_tensor(x, scale):
    assert scale > 0
    scale_matrix = get_scale_mat(scale, x.device, x.dtype)[None, ...].repeat(x.shape[0],1,1)                                        
    grid = F.affine_grid(scale_matrix, x.size())
    x = F.grid_sample(x, grid)
    return x