GAN실습 - 환경구축 & BeautyGAN(2)

이 자료는 한국인공지능협회에서 진행한 2021년 사업주 직업능력 개발훈련으로 BASIC AI(32시간)-박성주 강사의 온라인 교육내용을 참고한 것입니다. 

2021.12.16 - [그냥, AI] - GAN실습 - 환경구축 & BeautyGAN(1)

GAN실습 - 환경구축 & BeautyGAN(1)

 

# tensorflow의 낮은 버전으로 아래와 같은 오류가 나올 수 있으나 그냥 무시하면 된다

In [1]:

import dlib
import matplotlib.pyplot as plt
import matplotlib.patches as patches
import tensorflow as tf
import numpy as np

 

C:\ProgramData\Anaconda3\envs\p36t19_beauty\lib\site-packages\tensorflow\python\framework\dtypes.py:523: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint8 = np.dtype([("qint8", np.int8, 1)])
C:\ProgramData\Anaconda3\envs\p36t19_beauty\lib\site-packages\tensorflow\python\framework\dtypes.py:524: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
C:\ProgramData\Anaconda3\envs\p36t19_beauty\lib\site-packages\tensorflow\python\framework\dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint16 = np.dtype([("qint16", np.int16, 1)])
C:\ProgramData\Anaconda3\envs\p36t19_beauty\lib\site-packages\tensorflow\python\framework\dtypes.py:526: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
C:\ProgramData\Anaconda3\envs\p36t19_beauty\lib\site-packages\tensorflow\python\framework\dtypes.py:527: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint32 = np.dtype([("qint32", np.int32, 1)])
C:\ProgramData\Anaconda3\envs\p36t19_beauty\lib\site-packages\tensorflow\python\framework\dtypes.py:532: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  np_resource = np.dtype([("resource", np.ubyte, 1)])

 

Load Models¶

In [2]:

detector = dlib.get_frontal_face_detector()
sp = dlib.shape_predictor('models/shape_predictor_5_face_landmarks.dat')

 

Load Images¶

In [32]:

img = dlib.load_rgb_image('imgs/test4.jpg')

plt.figure(figsize=(16, 10))
plt.imshow(img)

Out[32]:

<matplotlib.image.AxesImage at 0x1f202564b38>

 

 

Find Faces¶

In [33]:

img_result = img.copy()

dets = detector(img, 1)

if len(dets) == 0:
    print('cannot find faces!')

fig, ax = plt.subplots(1, figsize=(16, 10))

for det in dets:
    x, y, w, h = det.left(), det.top(), det.width(), det.height()

    rect = patches.Rectangle((x, y), w, h, linewidth=2, edgecolor='r', facecolor='none')
    ax.add_patch(rect)

ax.imshow(img_result)

Out[33]:

<matplotlib.image.AxesImage at 0x1f201516780>

 

 

Find Landmarks 5points¶

In [34]:

fig, ax = plt.subplots(1, figsize=(16, 10))

objs = dlib.full_object_detections()

for detection in dets:
    s = sp(img, detection)
    objs.append(s)
    
    for point in s.parts():
        circle = patches.Circle((point.x, point.y), radius=3, edgecolor='r', facecolor='r')
        ax.add_patch(circle)

ax.imshow(img_result)

Out[34]:

<matplotlib.image.AxesImage at 0x1f2019ab1d0>

 

 

Align Faces¶

In [35]:

faces = dlib.get_face_chips(img, objs, size=256, padding=0.3)

fig, axes = plt.subplots(1, len(faces)+1, figsize=(20, 16))

axes[0].imshow(img)

for i, face in enumerate(faces):
    axes[i+1].imshow(face)

 

 

Functionalize¶

In [38]:

def align_faces(img):
    dets = detector(img, 1)
    
    objs = dlib.full_object_detections()

    for detection in dets:
        s = sp(img, detection)
        objs.append(s)
        
    faces = dlib.get_face_chips(img, objs, size=256, padding=0.35)
    
    return faces

# test
test_img = dlib.load_rgb_image('imgs/test5.jpg')

test_faces = align_faces(test_img)

fig, axes = plt.subplots(1, len(test_faces)+1, figsize=(20, 16))
axes[0].imshow(test_img)

for i, face in enumerate(test_faces):
    axes[i+1].imshow(face)

 

 

Load BeautyGAN Pretrained¶

• https://drive.google.com/drive/folders/1pgVqnF2-rnOxcUQ3SO4JwHUFTdiSe5t9

In [8]:

sess = tf.Session()
sess.run(tf.global_variables_initializer())

saver = tf.train.import_meta_graph('models/model.meta')
saver.restore(sess, tf.train.latest_checkpoint('models'))
graph = tf.get_default_graph()

X = graph.get_tensor_by_name('X:0') # source
Y = graph.get_tensor_by_name('Y:0') # reference
Xs = graph.get_tensor_by_name('generator/xs:0') # output

 

INFO:tensorflow:Restoring parameters from models\model

 

Preprocess and Postprocess Functions¶

In [9]:

def preprocess(img):
    return img.astype(np.float32) / 127.5 - 1.

def postprocess(img):
    return ((img + 1.) * 127.5).astype(np.uint8)

 

Load Images¶

In [40]:

img1 = dlib.load_rgb_image('imgs/test2.jpg')
img1_faces = align_faces(img1)

img2 = dlib.load_rgb_image('imgs/makeup/jolly.png')
img2_faces = align_faces(img2)

fig, axes = plt.subplots(1, 2, figsize=(16, 10))
axes[0].imshow(img1_faces[0])
axes[1].imshow(img2_faces[0])

Out[40]:

<matplotlib.image.AxesImage at 0x1f2035e5ac8>

 

 

Run¶

In [41]:

src_img = img1_faces[0]
ref_img = img2_faces[0]

X_img = preprocess(src_img)
X_img = np.expand_dims(X_img, axis=0)

Y_img = preprocess(ref_img)
Y_img = np.expand_dims(Y_img, axis=0)

output = sess.run(Xs, feed_dict={
    X: X_img,
    Y: Y_img
})

output_img = postprocess(output[0])

fig, axes = plt.subplots(1, 3, figsize=(20, 10))
axes[0].set_title('Source')
axes[0].imshow(src_img)
axes[1].set_title('Reference')
axes[1].imshow(ref_img)
axes[2].set_title('Result')
axes[2].imshow(output_img)

Out[41]:

<matplotlib.image.AxesImage at 0x1f201e13710>

 

In [ ]: