deepak112

from numpy import array
from scipy.misc import imresize
# Takes list of images and provide HR images in form of numpy array
def hr_images(images):
    images_hr = array(images)
    return images_hr

# Takes list of images and provide LR images in form of numpy array
def lr_images(images_real , downscale):   
    images = []

deepak112

from keras.layers.core import Activation
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import UpSampling2D
from keras.layers import Input
from keras.layers.convolutional import Conv2D, Conv2DTranspose
from keras.models import Model
from keras.layers.advanced_activations import LeakyReLU, PReLU
from keras.layers import add

# Residual block

deepak112

from keras.layers import Dense
from keras.layers.core import Activation
from keras.layers.normalization import BatchNormalization
from keras.layers.core import Flatten
from keras.layers import Input
from keras.layers.convolutional import Conv2D
from keras.models import Model
from keras.layers.advanced_activations import LeakyReLU

def discriminator_block(model, filters, kernel_size, strides):

deepak112

from keras.applications.vgg19 import VGG19
import keras.backend as K
from keras.models import Model

def vgg_loss(self, y_true, y_pred):
    vgg19 = VGG19(include_top=False, weights='imagenet', input_shape=self.image_shape)
    vgg19.trainable = False
    for l in vgg19.layers:
        l.trainable = False
    model = Model(inputs=vgg19.input, outputs=vgg19.get_layer('block5_conv4').output)

deepak112

from keras.models import Model
from keras.layers import Input

def get_gan_network(discriminator, shape, generator, optimizer, vgg_loss):
    discriminator.trainable = False
    gan_input = Input(shape=shape)
    x = generator(gan_input)
    gan_output = discriminator(x)
    gan = Model(inputs=gan_input, outputs=[x,gan_output])
    gan.compile(loss=[vgg_loss, "binary_crossentropy"], loss_weights=[1., 1e-3], optimizer=optimizer)

deepak112

def get_optimizer():
    
    adam = Adam(lr=1E-4, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
    return adam

deepak112

from tqdm import tqdm
import numpy as np

# Better to use downscale factor as 4
downscale_factor = 4
# Remember to change image shape if you are having different size of images
image_shape = (384,384,3)

def train(epochs, batch_size, input_dir, output_dir, model_save_dir, number_of_images, train_test_ratio):
    # Loads training and test data

deepak112

import cv2

cv2.namedWindow("image")
cv2.setMouseCallback("image", get_mouse_points)
def get_mouse_points(event, x, y, flags, param):

    global mouse_pts
    if event == cv2.EVENT_LBUTTONDOWN:
        if len(mouse_pts) < 4:
            cv2.circle(image, (x, y), 5, (0, 0, 255), 10)

deepak112

import cv2
import numpy as np

weightsPath = "path to yolov3.weights"
configPath = "path to yolov3.cfg"
vid_path = "path to video"

net_yl = cv2.dnn.readNetFromDarknet(configPath, weightsPath)
ln = net_yl.getLayerNames()
ln1 = [ln[i[0] - 1] for i in net_yl.getUnconnectedOutLayers()]

deepak112

import cv2
import numpy as np

(H, W) = frame.shape[:2]
points = mouse_pts      
        
# Using first 4 points or coordinates for perspective transformation. The region marked by these 4 points are 
# considered ROI. This polygon shaped ROI is then warped into a rectangle which becomes the bird eye view. 
# This bird eye view then has the property property that points are distributed uniformally horizontally and 
# vertically(scale for horizontal and vertical direction will be different). So for bird eye view points are