astolfo_classifier.py

# -*- coding: utf-8 -*-
"""Astolfo Classifier.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1_3DdxTiDkR-wZ686vxmn8q1T8-fnk44m
"""

import pandas as pd
import numpy as np
import os
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches

import tensorflow as tf
import keras
from keras import layers
from keras.models import Model
from keras.models import Sequential
from keras.layers import Conv2D, MaxPool2D, Flatten, Dense, InputLayer, BatchNormalization, Dropout
from tensorflow.keras.preprocessing.image import img_to_array
from keras import metrics

from sklearn.utils import shuffle

import cv2
import os
from tqdm import tqdm
import re

from google.colab import drive
drive.mount('/content/drive')

def sorted_alphanumeric(data):  
    convert = lambda text: int(text) if text.isdigit() else text.lower()
    alphanum_key = lambda key: [convert(c) for c in re.split('([0-9]+)',key)]
    return sorted(data,key = alphanum_key)

# defining the size of the image
SIZE = 160
train_img = []
train_label = []
path = '/content/drive/My Drive/Images_all/Astolpho'
files = os.listdir(path)
files = sorted_alphanumeric(files)
for i in tqdm(files):    
    if i == '1000.jpg':
        break
    else:    
        img = cv2.imread(path + '/'+i,1) 
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # open cv reads images in BGR format so we have to convert it to RGB
        img = cv2.resize(img, (SIZE, SIZE)) #resizing image
        img = img.astype('float32') / 255.0
        train_img.append(img_to_array(img))
        train_label.append('Astolpho');

path = '/content/drive/My Drive/Images_all/Non_Astolfo'
files = os.listdir(path)
files = sorted_alphanumeric(files)
for i in tqdm(files):    
    if i == '1000.jpg':
        break
    else:    
        img = cv2.imread(path + '/'+i,1) 
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # open cv reads images in BGR format so we have to convert it to RGB
        img = cv2.resize(img, (SIZE, SIZE)) #resizing image
        img = img.astype('float32') / 255.0
        train_img.append(img_to_array(img))
        train_label.append('Non_Astolpho');

path = '/content/drive/My Drive/Images_all/Non_Anime'
files = os.listdir(path)
files = sorted_alphanumeric(files)
for i in tqdm(files):    
    if i == '1000.jpg':
        break
    else:    
        img = cv2.imread(path + '/'+i,1) 
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # open cv reads images in BGR format so we have to convert it to RGB
        img = cv2.resize(img, (SIZE, SIZE)) #resizing image
        img = img.astype('float32') / 255.0
        train_img.append(img_to_array(img))
        train_label.append('Non_Anime');

# defining function to plot images pair
def plot_images(color):
    plt.figure(figsize=(15,15))
    plt.subplot(1,3,1)
    plt.title('Astolpho', color = 'green', fontsize = 20)
    plt.imshow(color)
   
    plt.show()

for i in range(61,80):
     plot_images(train_img[i])

x_train = np.array(train_img)

y_train = []
for i in train_label:
  if i=="Astolpho":
    y_train.append([1,0,0])
  if i=="Non_Anime":
    y_train.append([0,1,0])
  if i=="Non_Astolpho":
    y_train.append([0,0,1])

y_train = np.array(y_train)

y_train

"""## Saving and Loading Images"""

x_train.tofile('/content/drive/My Drive/Images_all/x_train.csv', sep = ',')
y_train.tofile('/content/drive/My Drive/Images_all/y_train.csv', sep = ',')

x_train = np.genfromtxt('/content/drive/My Drive/Images_all/x_train.csv', delimiter=',')
y_train = np.genfromtxt('/content/drive/My Drive/Images_all/y_train.csv', delimiter=',')

x_train = np.array(x_train)
y_train = np.array(y_train)

"""## Model"""

# from sklearn.preprocessing import LabelEncoder

# le = LabelEncoder()
# le.fit(y_train)
# # encode
# y_train = le.transform(y_train)

# !pip install category_encoders

# import category_encoders as ce

# enc=ce.OneHotEncoder().fit(y_train.astype(str))
# y_train=enc.transform(y_train.astype(str))

# # tf.keras.utils.to_categorical(y_train, num_classes=2, dtype='float32')

x_train, y_train = shuffle(x_train, y_train)

y_train

x_train.shape

# build a sequential model
model = Sequential()


model.add(InputLayer(input_shape=(160, 160, 3)))
model.add(Dense(units=100, activation='relu'))

# 1st conv block
model.add(Conv2D(25, (5, 5), activation='relu', strides=(1, 1), padding='same'))
model.add(MaxPool2D(pool_size=(2, 2), padding='same'))
model.add(BatchNormalization())

# # ANN block
model.add(Flatten())
model.add(Dense(units=100, activation='relu'))
model.add(Dropout(0.1))
# output layer
model.add(Dense(units=3, activation='softmax'))

model.summary()

from keras.callbacks import EarlyStopping, ReduceLROnPlateau

# compile model
model.compile(loss='binary_crossentropy', optimizer=tf.keras.optimizers.Adam(learning_rate=0.001, name="Adam"), metrics=['accuracy'])
# fit on data for 30 epochs

reducelr = ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=5, min_lr=0.0001)
es = EarlyStopping(monitor='val_loss', patience=5, min_delta=0.0001)
history = model.fit(x_train, y_train, epochs=30, validation_split = 0.2, verbose=1, callbacks=[reducelr, es])

img = cv2.imread('/content/drive/My Drive/Images_all/Randoms/lonely-japanese-cherry.jpg') 
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # open cv reads images in BGR format so we have to convert it to RGB
img = cv2.resize(img, (SIZE, SIZE)) #resizing image
img = img.astype('float32') / 255.0
test_img = img_to_array(img)
test_img = np.array(test_img)

test_img = np.reshape(test_img,(-1,SIZE,SIZE,3))

test_img.shape

model.predict(test_img)

model.save('/content/drive/My Drive/Images_all/saved_model/my_model.h5')