ksivaman

for ii in range(1, steps+1):
    
    # get the features from your target image
    target_features = get_features(target, vgg)
    
    # the content loss
    content_loss = torch.mean((target_features['conv4_2'] - content_features['conv4_2'])**2)
    
    # the style loss
    # initialize the style loss to 0

ksivaman

# get content and style features only once before training
content_features = get_features(content, vgg)
style_features = get_features(style, vgg)

# calculate the gram matrices for each layer of our style representation
style_grams = {layer: gram_matrix(style_features[layer]) for layer in style_features}

#initialize the target image as the content image
target = content.clone().requires_grad_(True).to(device)

ksivaman

def gram_matrix(tensor):

    # get the batch_size, depth, height, and width of the Tensor
    _, d, h, w = tensor.size()
    
    # reshape so we're multiplying the features for each channel
    tensor = tensor.view(d, h * w)
    
    # calculate the gram matrix
    gram = torch.mm(tensor, tensor.t())

ksivaman

def gram_matrix(tensor):

    # get the batch_size, depth, height, and width of the Tensor
    _, d, h, w = tensor.size()
    
    # reshape so we're multiplying the features for each channel
    tensor = tensor.view(d, h * w)
    
    # calculate the gram matrix
    gram = torch.mm(tensor, tensor.t())

ksivaman

def get_features(image, model, layers=None):
    """ Run an image forward through a model and get the features for 
        a set of layers. Default layers are for VGGNet matching Gatys et al (2016)
    """
    
    ## Need the layers for the content and style representations of an image
    if layers is None:
        layers = {'0': 'conv1_1',
                  '5': 'conv2_1', 
                  '10': 'conv3_1', 

ksivaman

vgg = models.vgg19(pretrained=True).features

# freeze VGG params to avoid chanhe
for param in vgg.parameters():
    param.requires_grad_(False)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
vgg.to(device)

ksivaman

from PIL import Image
from io import BytesIO
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import time

import torch
import torch.optim as optim
import requests

ksivaman

def train(train_X, train_Y, epochs, lr, layers=[4, 5, 1], activate=['R', 'S']):
    # initiation of neural netowrk parameters
    params_w, params_b = init(layers)

    losses = []
    accuracies = []
    
    # performing calculations for subsequent iterations
    for i in range(epochs):
        # step forward

ksivaman

def param_updates(params_w, params_b, gradients, lr, layers=[4, 5, 1]):

    for index in range(len(layers) - 1):
        #gradient descent
        params_w["weight" + str(index + 1)] -= lr * gradients["d_weight" + str(index + 1)]        
        params_b["bias" + str(index + 1)] -= lr * gradients["d_bias" + str(index + 1)]

    return params_w, params_b

ksivaman

#binary cross entropy loss
def cross_entropy_loss(y_pred, train_Y):
    num_samples = y_pred.shape[1]
    cost = -1 / num_samples * (np.dot(train_Y, np.log(y_pred).T) + np.dot(1 - train_Y, np.log(1 - y_pred).T))
    return np.squeeze(cost)

#convert probabilities to class prediction with threshold 0.5
def get_class_from_probs(probabilities):
    class_ = np.copy(probabilities)
    class_[class_ > 0.5] = 1