Storyboard

>Model

ID:(1779, 0)

Description

>Top

If you are looking for a list of names of the directories

that are associated with the directory sample_directory

example: 'F:/go/face_scrapper/faces/samples/'

you can use the listdir command from the library < b>os:

import os

# definiciones
sample_directory = 'F:/go/face_scrapper/faces/samples/'

# obtener arreglo de directorios
classes = []
maincls = os.listdir(sample_directory)

for cls in maincls:
    # generar list de archivos
    file = os.listdir(sample_directory+'/'+cls+'/')
    # agregar subdirectorios si contienen imagenes
    if len(file) > 0:
        classes.append(cls)

# mostrar classes
print(classes)

Only include folders with images.

ID:(13748, 0)

Description

>Top

If you are looking for a list of names of the subdirectories

that are associated with a main directory sample_directory

example: 'F:/go/face_scrapper/faces/samples/'

you can use the command listdir from the library os:

import os

# definitions
sample_directory = 'F:/go/face_scrapper/faces/samples/'

# get directory array
classes = []
maincls = os.listdir(sample_directory)

for subcls in maincls:

    # get subdirectory array
    clslst = os.listdir(sample_directory+subcls+'/')
    for cls in clslst:
        file = os.listdir(sample_directory+subcls+'/'+cls+'/')
    
        # add subdirectories to the array
        if len(file) > 0:
            classes.append(cls)
            
# show classes
print(classes)

ID:(13781, 0)

Description

>Top

To ensure that the classes array does not contain class for each type, all elements must be verified to be unique:

import collections
print([item for item, count in collections.Counter(classes).items() if count > 1])
print('number of elements:'+str(len(classes)))

The class array must not contain doubles.

ID:(13782, 0)

Description

>Top

To run the models, the directories are created
- train
- validate
- test
with the subdirectories of each image of the same classification depending from the work directory work_directory

example: 'F:/go/face_scrapper/faces/work/'

import os

# define directories
work_directory = 'F:/go/face_scrapper/faces/work/'

train_path = work_directory+'train/'
validate_path = work_directory+'validate/'
test_path = work_directory+'test/'

# create subdirectories
for subcls in classes:
    # define train directory
    FOLDER_EXIST = os.path.isdir(train_path+subcls)
    # check and create
    if not FOLDER_EXIST: os.mkdir(train_path+subcls)
    # define validate directory
    FOLDER_EXIST = os.path.isdir(validate_path+subcls)
    # check and create
    if not FOLDER_EXIST: os.mkdir(validate_path+subcls)
    # define test directory
    FOLDER_EXIST = os.path.isdir(test_path+subcls)
    # check and create
    if not FOLDER_EXIST: os.mkdir(test_path+subcls)

ID:(13749, 0)

Description

>Top

Once the directories are generated, the images must be copied to each subdirectory. To do this, you have to go through the subdirectories that depend on the sample directory sample_directory

example 'F:/go/face_scrapper/faces/samples/'

and copy them to the working directories work_directory

example 'F:/go/face_scrapper/faces/work/'

in its versions train (train), validate (validate) and test (test).

The number to be copied must have the fractions
- 80% train
- 15% validate
- 5% test

import shutil

# definitions
sample_directory = 'F:/go/face_scrapper/faces/samples/'
work_directory = 'F:/go/face_scrapper/faces/work/'

train_path = work_directory+'train/'
validate_path = work_directory+'validate/'
test_path = work_directory+'test/'

for subcls in classes:

    # get subclasses
    subclslst = os.listdir(sample_directory+subcls)

    # numbers
    nlen = len(subclslst)
    if nlen > 0:
        nlen1 = round(nlen*0.8)
        if nlen1 == 0: nlen1 = 1
        nlen2 = round(nlen*0.15)
        if nlen2 == 0: nlen2 = 1
        nlen3 = round(nlen*0.05)
        if nlen3 == 0: nlen3 = 1
        if nlen1 + nlen2 + nlen3 > nlen:
            npos3 = nlen - 1
        else:
            npos3 = nlen1 + nlen2
        if nlen1 + nlen2 > nlen:
            npos1 = nlen - 1
            npos2 = nlen
        else:
            npos1 = nlen1
            npos2 = nlen1 + nlen2            
        print(cls+','+str(nlen1)+','+str(nlen2)+','+str(nlen3))
        # copy files
        for i in range(0,nlen1):
            shutil.copyfile(sample_directory+subcls+'/'+subclslst[i],train_path+subcls+'/'+subclslst[i])
        for i in range(npos1,npos2):
            shutil.copyfile(sample_directory+subcls+'/'+subclslst[i],validate_path+subcls+'/'+subclslst[i])
        for i in range(npos3,nlen):
            shutil.copyfile(sample_directory+subcls+'/'+subclslst[i],test_path+subcls+'/'+subclslst[i])

ID:(13750, 0)

Description

>Top

Once the directories are generated, the images must be copied to each subdirectory. To do this, you have to go through the subdirectories that depend on the sample directory sample_directory

example 'F:/go/face_scrapper/faces/samples/'

and copy them to the working directories work_directory

example 'F:/go/face_scrapper/faces/work/'

in its versions train (train), validate (validate) and test (test).

The number to be copied must have the fractions
- 80% train
- 15% validate
- 5% test

import shutil

# definitions
sample_directory = 'F:/go/face_scrapper/faces/samples/'
work_directory = 'F:/go/face_scrapper/faces/work/'

train_path = work_directory+'train/'
validate_path = work_directory+'validate/'
test_path = work_directory+'test/'

# get directory array
maincls = os.listdir(sample_directory)

for subcls in maincls:

    # get subclasses
    temp = os.listdir(sample_directory+subcls+'/')    

    for cls in temp:

        # file list
        filelst = os.listdir(sample_directory+subcls+'/'+cls+'/')

        # numbers
        nlen = len(filelst)
        if nlen > 0:
            nlen1 = round(nlen*0.8)
            if nlen1 == 0: nlen1 = 1
            nlen2 = round(nlen*0.15)
            if nlen2 == 0: nlen2 = 1
            nlen3 = round(nlen*0.05)
            if nlen3 == 0: nlen3 = 1
            if nlen1 + nlen2 + nlen3 > nlen:
                npos3 = nlen - 1
            else:
                npos3 = nlen1 + nlen2
            if nlen1 + nlen2 > nlen:
                npos1 = nlen - 1
                npos2 = nlen
            else:
                npos1 = nlen1
                npos2 = nlen1 + nlen2            

            print(cls+','+str(nlen1)+','+str(nlen2)+','+str(nlen3))

            # copy files
            for i in range(0,nlen1):
                shutil.copyfile(sample_directory+subcls+'/'+cls+'/'+filelst[i],train_path+cls+'/'+filelst[i])
            for i in range(npos1,npos2):
                shutil.copyfile(sample_directory+subcls+'/'+cls+'/'+filelst[i],validate_path+cls+'/'+filelst[i])
            for i in range(npos3,nlen):
                shutil.copyfile(sample_directory+subcls+'/'+cls+'/'+filelst[i],test_path+cls+'/'+filelst[i])

ID:(13783, 0)

Description

>Top

In order to train the model, a pre-processing is carried out in which
- the images are scaled to the defined size (it is scaled regardless of whether the original image is larger, smaller and / or not square).
- contrast is increased by working with a different color palette

import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.preprocessing.image import ImageDataGenerator

# definitions
work_directory = 'F:/go/face_scrapper/faces/work/'

train_path = work_directory+'train/'
validate_path = work_directory+'validate/'
test_path = work_directory+'test/'

# generate images
train_batches = ImageDataGenerator(preprocessing_function=tf.keras.applications.vgg16.preprocess_input).flow_from_directory(directory=train_path, target_size=(224,224), classes=classes, batch_size=10)
valid_batches = ImageDataGenerator(preprocessing_function=tf.keras.applications.vgg16.preprocess_input).flow_from_directory(directory=valid_path, target_size=(224,224), classes=classes, batch_size=10)
test_batches = ImageDataGenerator(preprocessing_function=tf.keras.applications.vgg16.preprocess_input).flow_from_directory(directory=test_path, target_size=(224,224), classes=classes, batch_size=10, shuffle=False)

If you train your own model, the size of the images (target_size) can assume any value. However, if a pre-trained model is used, the size must be adjusted to the images of the original model.

ID:(13751, 0)

Description

>Top

La función ImageDataGenerator genera archivos con el formato dado por:

{'image_data_generator': ,

'target_size': (224, 224),

'color_mode': 'rgb',

'data_format': 'channels_last',

'image_shape': (224, 224, 3),

'save_to_dir': None,

'save_prefix': '',

'save_format': 'png',

'interpolation': 'nearest',

'split': None,

'subset': None,

'directory': 'F:/go/face_scrapper/faces/work/test/',

'classes': array([ 0, 1, 2, ..., 288, 289]),

'class_mode': 'categorical',

'dtype': 'float32',

'samples': 663,

'num_classes': 290,

'class_indices': {'arraya-andi': 0, 'arraya-marielle': 1, ...,'wagner-mancho': 288, 'wagner-richard': 289},

'filenames': ['arraya-andi

eg087_002_003_603-625-293-375.jpg',

...,

'wagner-richard\\sony-cybershot-20011204-1806002_001_201-163-141-172.jpg'],

'_filepaths': ['F:/go/face_scrapper/faces/work/test/arraya-andi

eg087_002_003_603-625-293-375.jpg',

...,

'F:/go/face_scrapper/faces/work/test/wagner-richard\\sony-cybershot-20011204-1806002_001_201-163-141-172.jpg'],

'n': 663,

'batch_size': 10,

'seed': None,

'shuffle': False,

'batch_index': 4,

'total_batches_seen': 140,

'lock': ,

'index_array': array([ 0, 1, 2, ..., 661, 662]),

'index_generator': }

test_

ID:(13799, 0)