CV之FD&FA：利用MTCNN的脚本实现对LFW数据集进行FD人脸检测和FA人脸校准

CV之FD&FA：利用MTCNN的脚本实现对LFW数据集进行FD人脸检测和FA人脸校准

目录

运行结果

运行过程

运行(部分)代码

在裁剪好的LFW数据集进行验证

运行结果

运行过程

time：14：03~14：37

usage: git diff [--no-index] <path> <path>
Creating networks and loading parameters
2018-10-01 14:03:00.211213: I tensorflow-regexp">/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that -variable language_">this TensorFlow binary was not compiled to use: AVX2
WARNING:tensorflow:From F:\AI\DL21TF\DL21examples\chapter_6\src\align\detect_face.py:210: calling reduce_max (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.
Instructions for updating:
keep_dims is deprecated, use keepdims instead
WARNING:tensorflow:From F:\AI\DL21TF\DL21examples\chapter_6\src\align\detect_face.py:212: calling reduce_sum (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.
Instructions for updating:
keep_dims is deprecated, use keepdims instead
datasets-regexp">/lfw/raw\Shane_Phillips\Shane_Phillips_0001.jpg
F:\Program Files\Python\Python36\Lib\site-packages\scipy\misc\pilutil.py:482: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int32 == np.dtype(int).type`.
  if issubdtype(ts, int):
F:\Program Files\Python\Python36\Lib\site-packages\scipy\misc\pilutil.py:485: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  elif issubdtype(type(size), float):
datasets-regexp">/lfw/raw\Alejandro_Lopez\Alejandro_Lopez_0001.jpg
……
 
 
datasets-regexp">/lfw/raw\Recep_Tayyip_Erdogan\Recep_Tayyip_Erdogan_0010.jpg
 
……
 
datasets-regexp">/lfw/raw\Ruth_Pearce\Ruth_Pearce_0001.jpg
datasets-regexp">/lfw/raw\Claudio_Abbado\Claudio_Abbado_0001.jpg
datasets-regexp">/lfw/raw\Dean_Jacek\Dean_Jacek_0001.jpg
datasets-regexp">/lfw/raw\Mark_Leno\Mark_Leno_0001.jpg
datasets-regexp">/lfw/raw\Kirk_Franklin\Kirk_Franklin_0001.jpg
datasets-regexp">/lfw/raw\Doris_Schroeder\Doris_Schroeder_0004.jpg
datasets-regexp">/lfw/raw\Doris_Schroeder\Doris_Schroeder_0002.jpg
datasets-regexp">/lfw/raw\Doris_Schroeder\Doris_Schroeder_0001.jpg
datasets-regexp">/lfw/raw\Doris_Schroeder\Doris_Schroeder_0003.jpg
datasets-regexp">/lfw/raw\Anne_Heche\Anne_Heche_0001.jpg
datasets-regexp">/lfw/raw\Keith_Fotta\Keith_Fotta_0001.jpg
datasets-regexp">/lfw/raw\Zdravko_Mucic\Zdravko_Mucic_0001.jpg
datasets-regexp">/lfw/raw\Mohammad_Fares\Mohammad_Fares_0001.jpg
datasets-regexp">/lfw/raw\Bruce_Lunsford\Bruce_Lunsford_0001.jpg
datasets-regexp">/lfw/raw\Craig_David\Craig_David_0001.jpg
datasets-regexp">/lfw/raw\Christine_Todd_Whitman\Christine_Todd_Whitman_0003.jpg
datasets-regexp">/lfw/raw\Christine_Todd_Whitman\Christine_Todd_Whitman_0006.jpg
datasets-regexp">/lfw/raw\Christine_Todd_Whitman\Christine_Todd_Whitman_0002.jpg
datasets-regexp">/lfw/raw\Christine_Todd_Whitman\Christine_Todd_Whitman_0001.jpg
datasets-regexp">/lfw/raw\Christine_Todd_Whitman\Christine_Todd_Whitman_0004.jpg
datasets-regexp">/lfw/raw\Christine_Todd_Whitman\Christine_Todd_Whitman_0005.jpg
datasets-regexp">/lfw/raw\Maritza_Macias_Furano\Maritza_Macias_Furano_0001.jpg
datasets-regexp">/lfw/raw\Edward_Burns\Edward_Burns_0001.jpg
datasets-regexp">/lfw/raw\Jeff_Roehm\Jeff_Roehm_0001.jpg
datasets-regexp">/lfw/raw\Mary_Frances_Seiter\Mary_Frances_Seiter_0001.jpg
datasets-regexp">/lfw/raw\Roger_Moore\Roger_Moore_0002.jpg
datasets-regexp">/lfw/raw\Roger_Moore\Roger_Moore_0003.jpg
datasets-regexp">/lfw/raw\Roger_Moore\Roger_Moore_0001.jpg
datasets-regexp">/lfw/raw\Roger_Moore\Roger_Moore_0005.jpg
datasets-regexp">/lfw/raw\Roger_Moore\Roger_Moore_0004.jpg
datasets-regexp">/lfw/raw\Jason_Priestley\Jason_Priestley_0001.jpg
datasets-regexp">/lfw/raw\Juergen_Chrobog\Juergen_Chrobog_0001.jpg
datasets-regexp">/lfw/raw\Deb_Santos\Deb_Santos_0001.jpg
datasets-regexp">/lfw/raw\Javier_Vazquez\Javier_Vazquez_0001.jpg
datasets-regexp">/lfw/raw\Vanessa_Laine\Vanessa_Laine_0001.jpg
datasets-regexp">/lfw/raw\Ilan_Goldfajn\Ilan_Goldfajn_0001.jpg
datasets-regexp">/lfw/raw\Tom_Osborne\Tom_Osborne_0001.jpg
datasets-regexp">/lfw/raw\Colleen_Donovan\Colleen_Donovan_0001.jpg
datasets-regexp">/lfw/raw\Jeff_Feldman\Jeff_Feldman_0001.jpg
datasets-regexp">/lfw/raw\DAngelo_Jimenez\DAngelo_Jimenez_0001.jpg
datasets-regexp">/lfw/raw\Jane_Riley\Jane_Riley_0001.jpg
datasets-regexp">/lfw/raw\Frank_Keating\Frank_Keating_0001.jpg
datasets-regexp">/lfw/raw\Rick_Romley\Rick_Romley_0001.jpg
datasets-regexp">/lfw/raw\Rick_Romley\Rick_Romley_0003.jpg
datasets-regexp">/lfw/raw\Rick_Romley\Rick_Romley_0002.jpg
datasets-regexp">/lfw/raw\Alexandre_Daigle\Alexandre_Daigle_0001.jpg
datasets-regexp">/lfw/raw\Stephen_Funk\Stephen_Funk_0001.jpg
datasets-regexp">/lfw/raw\Joe_Nichols\Joe_Nichols_0001.jpg
datasets-regexp">/lfw/raw\Joe_Nichols\Joe_Nichols_0003.jpg
datasets-regexp">/lfw/raw\Joe_Nichols\Joe_Nichols_0004.jpg
datasets-regexp">/lfw/raw\Joe_Nichols\Joe_Nichols_0002.jpg
datasets-regexp">/lfw/raw\Garth_Drabinsky\Garth_Drabinsky_0001.jpg
datasets-regexp">/lfw/raw\Pascal_Lamy\Pascal_Lamy_0001.jpg
datasets-regexp">/lfw/raw\Pascal_Lamy\Pascal_Lamy_0002.jpg
datasets-regexp">/lfw/raw\Raja_Ramani\Raja_Ramani_0001.jpg
datasets-regexp">/lfw/raw\David_Zeplowitz\David_Zeplowitz_0001.jpg
datasets-regexp">/lfw/raw\Kobe_Bryant\Kobe_Bryant_0003.jpg
datasets-regexp">/lfw/raw\Kobe_Bryant\Kobe_Bryant_0002.jpg
datasets-regexp">/lfw/raw\Kobe_Bryant\Kobe_Bryant_0001.jpg
 
……
 
datasets-regexp">/lfw/raw\Renee_Zellweger\Renee_Zellweger_0002.jpg
datasets-regexp">/lfw/raw\Renee_Zellweger\Renee_Zellweger_0004.jpg
datasets-regexp">/lfw/raw\Renee_Zellweger\Renee_Zellweger_0015.jpg
datasets-regexp">/lfw/raw\Jack_Nicholson\Jack_Nicholson_0003.jpg
datasets-regexp">/lfw/raw\Jack_Nicholson\Jack_Nicholson_0001.jpg
datasets-regexp">/lfw/raw\Jack_Nicholson\Jack_Nicholson_0002.jpg
datasets-regexp">/lfw/raw\Edward_Egan\Edward_Egan_0001.jpg
datasets-regexp">/lfw/raw\Ion_Tiriac\Ion_Tiriac_0001.jpg
datasets-regexp">/lfw/raw\Jack_Knowlton\Jack_Knowlton_0001.jpg
datasets-regexp">/lfw/raw\Akbar_Al_Baker\Akbar_Al_Baker_0001.jpg
datasets-regexp">/lfw/raw\Gholamreza_Aghazadeh\Gholamreza_Aghazadeh_0001.jpg
datasets-regexp">/lfw/raw\Ramon_Cardenas\Ramon_Cardenas_0001.jpg
datasets-regexp">/lfw/raw\Ted_Maher\Ted_Maher_0002.jpg
datasets-regexp">/lfw/raw\Ted_Maher\Ted_Maher_0001.jpg
datasets-regexp">/lfw/raw\Dionyssis_Georgiadis\Dionyssis_Georgiadis_0001.jpg
datasets-regexp">/lfw/raw\Gennifer_Flowers\Gennifer_Flowers_0001.jpg
datasets-regexp">/lfw/raw\Bernard_Giraudeau\Bernard_Giraudeau_0001.jpg
datasets-regexp">/lfw/raw\Mary_Hill\Mary_Hill_0001.jpg
datasets-regexp">/lfw/raw\John_Leguizamo\John_Leguizamo_0001.jpg
datasets-regexp">/lfw/raw\Celia_Cruz\Celia_Cruz_0001.jpg
datasets-regexp">/lfw/raw\Igor_Ivanov\Igor_Ivanov_0001.jpg
 
……
 
datasets-regexp">/lfw/raw\Johnny_Benson\Johnny_Benson_0001.jpg
datasets-regexp">/lfw/raw\Ronaldo_Luis_Nazario_de_Lima\Ronaldo_Luis_Nazario_de_Lima_0004.jpg
datasets-regexp">/lfw/raw\Ronaldo_Luis_Nazario_de_Lima\Ronaldo_Luis_Nazario_de_Lima_0002.jpg
datasets-regexp">/lfw/raw\Ronaldo_Luis_Nazario_de_Lima\Ronaldo_Luis_Nazario_de_Lima_0003.jpg
datasets-regexp">/lfw/raw\Ronaldo_Luis_Nazario_de_Lima\Ronaldo_Luis_Nazario_de_Lima_0001.jpg
datasets-regexp">/lfw/raw\Carroll_Weimer\Carroll_Weimer_0001.jpg
datasets-regexp">/lfw/raw\Mark_Butcher\Mark_Butcher_0001.jpg
datasets-regexp">/lfw/raw\Takenori_Kanzaki\Takenori_Kanzaki_0001.jpg
Total number of images: 13233
Number of successfully aligned images: 13233

运行(部分)代码

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
 
from scipy import misc
import sys
import os
import argparse
import tensorflow as tf
 
import facenet
import align.detect_face
import random
from time import sleep
 
 
 
def parse_arguments(argv):
    parser = argparse.ArgumentParser()
 
    parser.add_argument('input_dir', type=str, help='Directory with unaligned images.')
    parser.add_argument('output_dir', type=str, help='Directory with aligned face thumbnails.')
    parser.add_argument('--image_size', type=int,
 
……
 
    return parser.parse_args(argv)
 
if __name__ == '__main__':
    main(parse_arguments(sys.argv[1:]))

在裁剪好的LFW数据集进行验证

Instructions for updating:
To construct input pipelines, use the `tf.data` module.
Runnning forward pass on LFW images
Accuracy: 0.992+-0.003
Validation rate: 0.97467+-0.01477 @ FAR=0.00133
Area Under Curve (AUC): 1.000
Equal Error Rate (EER): 0.007