!
pip install tensorflow tensorflow -gpu opencv-python matplotlib
import tensorflow as tf
import os # os is used for joining the data like connecting data_dir with image Class(happy)
# Avoid OOM errors by setting GPU Memory Consumption Growth
gpus = [Link].list_physical_devices( 'GPU')
for gpu in gpus:
[Link].set_memory_growth( gpu, True)
[Link].list_physical_devices('GPU')
# 2. Remove dodgy images
import cv2
import imghdr
read_image=[Link]([Link](x,y,z))-> [Link](image) *cv read in BGR while Matplot in RGB
[Link]([Link](image,cv2.COLOR_BGR2RGB ))
data_dir = 'data'
image_exts = ['jpeg','jpg', 'bmp', 'png']
for image_class in [Link](data_dir):
for image in [Link]([Link](data_dir, image_class)):
image_path = [Link](data_dir, image_class, image)
try:
img = [Link](image_path)
tip = [Link](image_path)
if tip not in image_exts:
print('Image not in ext list {}'.format(image_path))
[Link](image_path)
except Exception as e:
print('Issue with image {}'.format(image_path))
# [Link](image_path)
# 3. Load Data
import numpy as np
from matplotlib #[Link]
data=[Link].image_dataset_from_directory(dir/subfld ') [Link] READ
cons this don’t pre load in mem but we need to grab data using numpt iterator
data_iterator = data.as_numpy_iterator()
batch = data_iterator.next()
fig, ax = [Link](ncols=4, figsize=(20,20))
for idx, img in enumerate(batch[0][:4]):
ax[idx].imshow([Link](int))
ax[idx].title.set_text(batch[1][idx])
# 4. Scale Data
data = [Link](lambda x,y: (x/255, y))
data.as_numpy_iterator().next()
# 5. Split Data
train_size = int(len(data)*.7)
val_size = int(len(data)*.2)
test_size = int(len(data)*.1)
train_size
train = [Link](train_size)
val = [Link](train_size).take(val_size)
test = [Link](train_size+val_size).take(test_size)
# 6. Build Deep Learning Model
train
\from [Link] import Sequential
from [Link] import Conv2D, MaxPooling2D, Dense, Flatten, Dropout
model = Sequential()
[Link](Conv2D(16, (3,3), 1, activation='relu', input_shape=(256,256,3)))
[Link](MaxPooling2D())
[Link](Conv2D(32, (3,3), 1, activation='relu'))
[Link](MaxPooling2D())
[Link](Conv2D(16, (3,3), 1, activation='relu'))
[Link](MaxPooling2D())
[Link](Flatten())
[Link](Dense(256, activation='relu'))
[Link](Dense(1, activation='sigmoid'))
[Link]('adam', loss=[Link](), metrics=['accuracy'])
[Link]()
# 7. Train
logdir='logs'
tensorboard_callback = [Link](log_dir=logdir)
hist = [Link](train, epochs=20, validation_data=val,
callbacks=[tensorboard_callback])
# 8. Plot Performance
fig = [Link]()
[Link]([Link]['loss'], color='teal', label='loss')
[Link]([Link]['val_loss'], color='orange', label='val_loss')
[Link]('Loss', fontsize=20)
[Link](loc="upper left")
[Link]()
fig = [Link]()
[Link]([Link]['accuracy'], color='teal', label='accuracy')
[Link]([Link]['val_accuracy'], color='orange', label='val_accuracy')
[Link]('Accuracy', fontsize=20)
[Link](loc="upper left")
[Link]()
# 9. Evaluate
from [Link] import Precision, Recall, BinaryAccuracy
pre = Precision()
re = Recall()
acc = BinaryAccuracy()
for batch in test.as_numpy_iterator():
X, y = batch
yhat = [Link](X)
pre.update_state(y, yhat)
re.update_state(y, yhat)
acc.update_state(y, yhat)
print([Link](), [Link](), [Link]())
# 10. Test
import cv2
img = [Link]('[Link]')
[Link](img)
[Link]()
resize = [Link](img, (256,256))
[Link]([Link]().astype(int))
[Link]()
yhat = [Link](np.expand_dims(resize/255, 0))
yhat
if yhat > 0.5:
print(f'Predicted class is Sad')
else:
print(f'Predicted class is Happy')
# 11. Save the Model
from [Link] import load_model
[Link]([Link]('models','imageclassifier.h5'))
new_model = load_model('imageclassifier.h5')
new_model.predict(np.expand_dims(resize/255, 0))
