virtualdvid

import os
import gc
import time
import psutil
from keras.preprocessing.image import ImageDataGenerator
from gapcv.vision import Images

virtualdvid

MB = (1024 * 1024)
GB = (MB * 1024)
start_mem = psutil.virtual_memory().used
print("Start Memory: %.2f GB" % (start_mem / GB))

virtualdvid

# Image Preprocessing

start_time = time.time()

dataset = Images('flowers', 'flowers', config=['resize=(128,128)', 'store'])

print("Preprocessing: ", int(time.time() - start_time), "secs")
print("Classes", dataset.classes, "Images:", dataset.count)

curr_mem = psutil.virtual_memory().used

virtualdvid

STEPS = 1000

print("Keras: Flow")

datagen = ImageDataGenerator()

start_time = time.time()

datagen.fit(X_train)
train_generator = datagen.flow(X_train, Y_train, batch_size=32)

virtualdvid

STEPS = 1000

print("Gap: Flow")

start_mem = psutil.virtual_memory().used
start_time = time.time()

dataset.minibatch = 32
train_generator = dataset.minibatch
print("PRE-TIME", time.time() - start_time)

virtualdvid

virtualdvid

print("Keras: flow_from_directory")

start_mem = psutil.virtual_memory().used
start_time = time.time()

datagen = ImageDataGenerator(rescale=1./255)
train_generator = datagen.flow_from_directory('flowers', target_size=(128,128), batch_size=32)
print("PRE-TIME", time.time() - start_time)

start_time = time.time()

virtualdvid

print("Gap: flow from directory")

start_mem = psutil.virtual_memory().used
start_time = time.time()

dataset = Images(config=['stream'])
dataset.load('flowers')
dataset.minibatch = 32
train_generator = dataset.minibatch

virtualdvid

virtualdvid

print("Keras: Flow with Augmentation")

datagen = ImageDataGenerator(rotation_range=30)

start_time = time.time()

datagen.fit(X_train)
train_generator = datagen.flow(X_train, Y_train, batch_size=32)

print("PRE-TIME (fit)", time.time() - start_time)