WNOkerasalak

from keras.models import Sequential
from keras.layers import Conv2D
from keras.layers import Flatten, MaxPooling2D
from keras.models import Sequential
from keras.layers import Dense, Activation, Dropout
from keras.models import load_model
import numpy as np

import random
import cv2
import imutils
import glob

def dawaj():
    y_train = []
    y_validate = []
    size = (200, 200)
    train_range = 200
    random.seed()
    for i in range(100):
        a = random.randrange(1, 11)
        b = random.randrange(0, 9)
        image = cv2.imread('im{}.jpg'.format(a))
        sword = cv2.imread('mieczalak/miecz{}.jpg'.format(b))
        alfa = random.randrange(0, 360)
        sword = imutils.rotate_bound(sword, alfa)
        size2 = (int(np.shape(sword)[1]/2), int(np.shape(sword)[0]/2))
        x = random.randrange(0, size[1] - size2[1])
        y = random.randrange(0, size[0] - size2[0])
        image = cv2.resize(image, size)
        sword = cv2.resize(sword, size2)
        for j in range(size2[0]):
            for k in range(size2[1]):
                if sword[k, j, 0] == 0 and sword[k, j, 1] == 0 and sword[k, j, 2] == 0:
                    pass
                else:
                    image[x + k, y + j, 0] = sword[k, j, 0]
                    image[x + k, y + j, 1] = sword[k, j, 1]
                    image[x + k, y + j, 2] = sword[k, j, 2]
        if i < 80:
            cv2.imwrite('dataset/train/im{}.jpg'.format(i), image)
            y_train.append([(y + size2[0]/2)/size[0], (x + size2[1]/2)/size[1]])
        else:
            cv2.imwrite('dataset/validate/im{}.jpg'.format(i), image)
            y_validate.append([(y + size2[0]/2)/size[0], (x+size2[1]/2)/size[1]])

    x_train = [cv2.imread(file) for file in glob.glob('./dataset/train/*.jpg')]
    x_validate = [cv2.imread(file) for file in glob.glob('./dataset/validate/*.jpg')]
    x_train = np.array(x_train)
    x_validate = np.array(x_validate)
    x_train = (x_train)/255
    x_validate = (x_validate)/255
    y_train = np.array(y_train)
    y_validate = np.array(y_validate)
    return x_train, y_train, x_validate, y_validate, size


x_train, y_train, x_validate, y_validate, size = dawaj()

classifier = Sequential([
        Conv2D(32, (5, 5), input_shape=(size[0], size[1], 3), activation='relu'),
        MaxPooling2D(pool_size=(2, 2)),
        Conv2D(32, (5, 5), activation='sigmoid'),
        MaxPooling2D(pool_size=(2, 2)),
        Flatten(),
        Dense(units=128, activation='sigmoid'),
        Dropout(rate=0.4),
        Dense(units=2, activation='sigmoid')
])

classifier.compile('adadelta', 'mse', metrics=['accuracy'])
# classifier.compile('adam', 'mse', metrics=['accuracy'])
#classifier = load_model('model_lab3.h5')
classifier.fit(x_train, y_train, epochs=40, validation_data=(x_validate, y_validate))

classifier.save('model_lab3.h5')