Я пытаюсь создать бинарный классификатор изображений, чтобы определить, использует ли кто-то хирургическую маску по примеру «Кошка против собак», доступному на веб-сайте TensorFlow (https://www.tensorflow.org/tutorials/images/classification)
Я создал небольшой набор данных с несколькими изображениями людей, одетых в хирургические маски, а некоторых людей без них, обучил мою CNN и получил точность около 70%, что к настоящему времени хорошо. Но дело в том, как мне делать прогнозы? Пример 'CatVsDogs' останавливается на увеличении.
Сейчас я не беспокоюсь о точности, просто удивляюсь, как я могу получить прогнозы из моей модели.
Это мой код:
import sys
import time
import numpy as np
import matplotlib.pyplot as plt
import os
import cv2
import random
import keras
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten, Conv2D, MaxPooling2D
from tensorflow.keras.callbacks import TensorBoard
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import adam
IMG_SIZE = 100 # Image dimensions
batch_size = 100 # Amount of data that will be fed to the NN at a time
epochs = 1 # Amount of times the data will be pass to the NN
training_data = []
#img_array = []
new_array = []
####### DATASET LOCATION #######
TRAIN_DIR = 'C:/Users/Alex/Google Drive/Colab Notebooks/MaskDetector/Train/' #Create variable to store the path of the training images directory
VALIDATION_DIR = 'C:/Users/Alex/Google Drive/Colab Notebooks/MaskDetector/Validate/' #Create variable to store the path of the validation images directory
TEST_DIR = 'C:/Users/Alex/Google Drive/Colab Notebooks/MaskDetector/Test/' #Create variable to store the path of the testing images directory
CATEGORIES = ['MaskOn','MaskOff'] #Categories 'MaskOn' and 'MaskOff' same name as the folders
TRAIN_DIR_MASKON = os.path.join(TRAIN_DIR, 'MaskOn') # Directory with pics of ppl with masks for training
TRAIN_DIR_MASKOFF = os.path.join(TRAIN_DIR, 'MaskOff') # Directory with pics of ppl without masks for training
VALIDATION_DIR_MASKON = os.path.join(VALIDATION_DIR, 'MaskOn') # Directory with pics of ppl with masks for validation
VALIDATION_DIR_MASKOFF = os.path.join(VALIDATION_DIR, 'MaskOff') # Directory with pics of ppl without masks for validation
###########################################
####### Shows the size of the dataset #######
num_maskon_tr = len(os.listdir(TRAIN_DIR_MASKON))
num_maskoff_tr = len(os.listdir(TRAIN_DIR_MASKOFF))
num_maskon_val = len(os.listdir(VALIDATION_DIR_MASKON))
num_maskoff_val = len(os.listdir(VALIDATION_DIR_MASKOFF))
total_train = num_maskon_tr + num_maskoff_tr
total_val = num_maskon_val + num_maskoff_val
###################### DATA AUGMENTATION ######################
######### FLIP #########
image_gen = ImageDataGenerator(rescale=1./255, horizontal_flip=True)
train_data_gen = image_gen.flow_from_directory(batch_size=batch_size,
directory=TRAIN_DIR,
shuffle=True,
target_size=(IMG_SIZE, IMG_SIZE))
augmented_images = [train_data_gen[0][0][0] for i in range(5)]
# Re-use the same custom plotting function defined and used above to visualize the training images
######### ROTATE 45° #########
image_gen = ImageDataGenerator(rescale=1./255, rotation_range=45)
train_data_gen = image_gen.flow_from_directory(batch_size=batch_size,
directory=TRAIN_DIR,
shuffle=True,
target_size=(IMG_SIZE, IMG_SIZE))
augmented_images = [train_data_gen[0][0][0] for i in range(5)]
######### ZOOM FROM 0 TO 10% #########
# zoom_range from 0 - 1 where 1 = 100%.
image_gen = ImageDataGenerator(rescale=1./255, zoom_range=0.5) #
train_data_gen = image_gen.flow_from_directory(batch_size=batch_size,
directory=TRAIN_DIR,
shuffle=True,
target_size=(IMG_SIZE, IMG_SIZE))
augmented_images = [train_data_gen[0][0][0] for i in range(5)]
####################################################################################################
### PREPARES THE DATA TO BE FED INTO THE NN ###
train_image_generator = ImageDataGenerator(rescale=1./255)
image_gen_val = ImageDataGenerator(rescale=1./255)
test_data_generator = ImageDataGenerator(rescale=1./255) #Use this to make predictons?Don't know yet
val_data_gen = image_gen_val.flow_from_directory(batch_size=batch_size,
directory=VALIDATION_DIR,
target_size=(IMG_SIZE, IMG_SIZE),
class_mode='binary')
####################################################################################
train_data_gen = train_image_generator.flow_from_directory(batch_size=batch_size,
directory=TRAIN_DIR,
shuffle=True,
target_size=(IMG_SIZE, IMG_SIZE),
class_mode='binary')
#####################################################################################
test_generator = test_data_generator.flow_from_directory(TEST_DIR,
target_size=(IMG_SIZE, IMG_SIZE),
batch_size=batch_size,
class_mode="binary",
shuffle=True)
############ NN Model ############
model = Sequential([
Conv2D(16, 3, padding='same', activation='relu',
input_shape=(IMG_SIZE, IMG_SIZE ,3)),
MaxPooling2D(),
Dropout(0.2),
Conv2D(32, 3, padding='same', activation='relu'),
MaxPooling2D(),
Conv2D(64, 3, padding='same', activation='relu'),
MaxPooling2D(),
Dropout(0.2),
Flatten(),
Dense(512, activation='relu'),
Dense(1)
])
#########################################################
############ COMPILES THE NN ############
model.compile(optimizer='adam',
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=['accuracy'])
############ PERFORMS THE TRAINNING ############
history = model.fit_generator(
train_data_gen,
steps_per_epoch=total_train // batch_size,
epochs=epochs,
validation_data=val_data_gen,
validation_steps=total_val // batch_size
)