The project consist in applying Style Transfer from one picture to another.
The content image was taken from the window of a building in Santos/SP - Brazil. In this picture, it is possible to see parts of the city, the port, and parts of Guarujá which is the neighboring city.
The style image is a painting from Anita Malfatti, who was a Brazillian artist, entitled "O farol" from 1915.
The algorithm is taken straight from class 6 - Style Transfer, from Udacity course.
The notebook was executed in Google Colaboratory. The first 3 cells denotes the first steps to take after importing the notebook to Colab.
The cell bellow will install the latest stable version of PyTorch
# http://pytorch.org/
from os.path import exists
from wheel.pep425tags import get_abbr_impl, get_impl_ver, get_abi_tag
platform = '{}{}-{}'.format(get_abbr_impl(), get_impl_ver(), get_abi_tag())
cuda_output = !ldconfig -p|grep cudart.so|sed -e 's/.*\.\([0-9]*\)\.\([0-9]*\)$/cu\1\2/'
accelerator = cuda_output[0] if exists('/dev/nvidia0') else 'cpu'
!pip install -q torch torchvision
Then the PIL package is updated to avoid compatibility errors.
!pip install PIL
The third cell is necessary only if you desire to save your work directly to Google Drive. I followed the steps in this article:
from google.colab import drive
drive.mount('/content/gdrive')
The following cells will apply the script from the class and generate a file named 'target.jpg' with the target image.