Computer Vision News - January 2020

3 Summary Generating High Fidelity I ages ... 7 Our output image is just the variable target. Using the utility function tensor_to_ image we can transform it back to an image and draw the results. As can be seen, we got what we expected: an image where the content is the shoes, and the style of the image is similar to the abstract painting above: PyTorch is an end to end library for implementing deep learning models. It allows flexibility for the user while keeping simplicity, which makes it the perfect tool for deeplearningresearch. Inthearticle,wesawanexample of neural style transfer which demonstrates from one side the simple syntax of PyTorch, and from the other handtheeaseofmanipulatingthenetworkcomponents. We saw many useful features such as weights freezing, extraction of intermediate features from the network, basic training and more. Today PyTorch is the most Conclusion We highly recommend you try the code yourself and by doing that learn some advanced tools of PyTorch common tool in academia and becoming increasingly popular in the industry. We highly recommend you try the code yourself and by doing that learn some advanced tools of PyTorch. Enjoy! Below is the utility function we used as well as all the imports that are needed. from torchvision import models from torchvision import transforms as tf import torch from PIL import Image import torch.optim as optim import matplotlib.pyplot as plt import numpy as np import os def transformation (img): tasks = tf . Compose([tf . Resize( 400 ), tf . ToTensor(), tf . Normalize(( 0.44 , 0.44 , 0.44 ),( 0.22 , 0.22 , 0.22 ))]) img = tasks(img)[: 3 ,:,:] . unsqueeze( 0 ) return img def correlation_matrix (tensor): _, d, h, w = tensor . size() tensor = tensor . view(d, h * w) correlation = torch . mm(tensor, tensor . t()) return correlation def tensor_to_image (tensor): image = tensor . to( "cpu" ) . clone() . detach() image = image . numpy() . squeeze() image = image . transpose( 1 , 2 , 0 ) image *= np . array(( 0.22 , 0.22 , 0.22 )) + np . array(( 0.44 , 0.44 , 0.44 )) image = image . clip( 0 , 1 ) return image