Hands-on GANs in practice

Let’s examine how we can utilize a GAN to generate some synthetic images in practice. We will examine Closed-Form Factorization of Latent Semantics in GANs (https://arxiv.org/abs/2007.06600) to learn how we can simply generate synthetic images for our ML problem. The code for this example was adapted from the paper’s original GitHub (https://github.com/genforce/sefa).

We begin by importing the essential libraries as shown:

# import the required libraries
import cv2
import torch
import numpy as np
from utils import to_tensor
from utils import postprocess
from utils import load_generator
from models import parse_gan_type
from utils import factorize_weight
from matplotlib import pyplot as plt

Then, we select the parameters of the generation process such as the number of images to generate, and the noise seed. Please note that the seed parameter will help us to get diverse images in this example:

num_samples = 1 # num of image to generate...