├── .github
└── FUNDING.yml
├── AnimeGAN.py
├── README.md
├── Requirements installation tutorial.md
├── checkpoint
├── AnimeGAN_Hayao_lsgan_300_300_1_2_10
│ └── checkpoints files
└── generator_Hayao_weight
│ ├── Hayao-60.ckpt.data-00000-of-00001
│ ├── Hayao-60.ckpt.index
│ ├── Hayao-60.ckpt.meta
│ └── checkpoint
├── dataset
├── Hayao
│ ├── smooth
│ │ └── smooth dir
│ └── style
│ │ └── style images dir
├── Paprika
│ ├── smooth dir
│ │ └── smooth images
│ └── style
│ │ └── style images dir
├── Shinkai
│ └── sytle
│ │ └── sytle images dir
├── test
│ ├── HR_photo
│ │ └── HR photos
│ ├── real
│ │ └── photos
│ ├── test_photo
│ │ └── none
│ └── test_photo256
│ │ └── 256x256 photos
├── train_photo
│ └── train photos
└── val
│ └── valid images
├── doc
├── 3-1block.png
├── Chen2020_Chapter_AnimeGAN.pdf
├── GD.png
├── Japanese_README.md
├── e2.png
├── e3.png
├── e4.png
└── sota.png
├── net
├── discriminator.py
└── generator.py
├── result
└── Hayao
│ ├── SOTA_comparison
│ ├── AnimeGAN
│ │ ├── 105.jpg
│ │ ├── 185.jpg
│ │ ├── 19.jpg
│ │ ├── 190.jpg
│ │ ├── 198.jpg
│ │ ├── 204.jpg
│ │ ├── 208.jpg
│ │ ├── 246.jpg
│ │ ├── 256.jpg
│ │ ├── 258.jpg
│ │ ├── 26.jpg
│ │ ├── 35.jpg
│ │ ├── 447.jpg
│ │ ├── 539.jpg
│ │ ├── 599.jpg
│ │ ├── 618.jpg
│ │ ├── 632.jpg
│ │ ├── 675.jpg
│ │ ├── 681.jpg
│ │ ├── 69.jpg
│ │ ├── 75.jpg
│ │ └── 762.jpg
│ ├── CartoonGAN
│ │ ├── 105_1562736508.jpg
│ │ ├── 185_1562736844.jpg
│ │ ├── 190_1562737136.jpg
│ │ ├── 198_1562737236.jpg
│ │ ├── 19_1562736964.jpg
│ │ ├── 204_1562737290.jpg
│ │ ├── 208_1562737392.jpg
│ │ ├── 246_1562737460.jpg
│ │ ├── 256_1562739338.jpg
│ │ ├── 258_1562737714.jpg
│ │ ├── 26_1562737803.jpg
│ │ ├── 35_1562737865.jpg
│ │ ├── 447_1562737943.jpg
│ │ ├── 539_1562738321.jpg
│ │ ├── 599_1562739505.jpg
│ │ ├── 618_1562739199.jpg
│ │ ├── 632_1562738596.jpg
│ │ ├── 675_1562738662.jpg
│ │ ├── 681_1562738765.jpg
│ │ ├── 69_1562738956.jpg
│ │ ├── 75_1562739149.jpg
│ │ └── 762_1562739095.jpg
│ ├── ComixGAN
│ │ ├── 105.png
│ │ ├── 185.png
│ │ ├── 19.png
│ │ ├── 190.png
│ │ ├── 198.png
│ │ ├── 204.png
│ │ ├── 208.png
│ │ ├── 246.png
│ │ ├── 256.png
│ │ ├── 258.png
│ │ ├── 26.png
│ │ ├── 35.png
│ │ ├── 447.png
│ │ ├── 539.png
│ │ ├── 599.png
│ │ ├── 618.png
│ │ ├── 632.png
│ │ ├── 675.png
│ │ ├── 681.png
│ │ ├── 69.png
│ │ ├── 75.png
│ │ └── 762.png
│ └── real
│ │ ├── 105.jpg
│ │ ├── 185.jpg
│ │ ├── 19.jpg
│ │ ├── 190.jpg
│ │ ├── 198.jpg
│ │ ├── 204.jpg
│ │ ├── 208.jpg
│ │ ├── 246.jpg
│ │ ├── 256.jpg
│ │ ├── 258.jpg
│ │ ├── 26.jpg
│ │ ├── 35.jpg
│ │ ├── 447.jpg
│ │ ├── 539.jpg
│ │ ├── 599.jpg
│ │ ├── 618.jpg
│ │ ├── 632.jpg
│ │ ├── 675.jpg
│ │ ├── 681.jpg
│ │ ├── 69.jpg
│ │ ├── 75.jpg
│ │ └── 762.jpg
│ ├── photo
│ ├── 1.jpg
│ ├── 10.jpg
│ ├── 11.jpg
│ ├── 12.jpg
│ ├── 13.jpg
│ ├── 14.jpg
│ ├── 15.jpg
│ ├── 16.jpg
│ ├── 17.jpg
│ ├── 18.jpg
│ ├── 19.jpg
│ ├── 2.jpg
│ ├── 20.jpg
│ ├── 21.jpg
│ ├── 22.jpg
│ ├── 23.jpg
│ ├── 24.jpg
│ ├── 25.jpg
│ ├── 26.jpg
│ ├── 27.jpg
│ ├── 28.jpg
│ ├── 29.jpg
│ ├── 3.jpg
│ ├── 30.jpg
│ ├── 31.jpg
│ ├── 32.jpg
│ ├── 33.jpg
│ ├── 34.jpg
│ ├── 35.jpg
│ ├── 36.jpg
│ ├── 37.jpg
│ ├── 38.jpg
│ ├── 39.jpg
│ ├── 4.jpg
│ ├── 40.jpg
│ ├── 41.jpg
│ ├── 42.jpg
│ ├── 43.jpg
│ ├── 44.jpg
│ ├── 45.jpg
│ ├── 5.jpg
│ ├── 6.jpg
│ ├── 7.jpg
│ ├── 8.jpg
│ ├── 9.jpg
│ └── AE86.png
│ └── photo_result
│ ├── 1.jpg
│ ├── 10.jpg
│ ├── 11.jpg
│ ├── 12.jpg
│ ├── 13.jpg
│ ├── 14.jpg
│ ├── 15.jpg
│ ├── 16.jpg
│ ├── 17.jpg
│ ├── 18.jpg
│ ├── 19.jpg
│ ├── 2.jpg
│ ├── 20.jpg
│ ├── 21.jpg
│ ├── 22.jpg
│ ├── 23.jpg
│ ├── 24.jpg
│ ├── 25.jpg
│ ├── 26.jpg
│ ├── 27.jpg
│ ├── 28.jpg
│ ├── 29.jpg
│ ├── 3.jpg
│ ├── 30.jpg
│ ├── 31.jpg
│ ├── 32.jpg
│ ├── 33.jpg
│ ├── 34.jpg
│ ├── 35.jpg
│ ├── 36.jpg
│ ├── 37.jpg
│ ├── 38.jpg
│ ├── 39.jpg
│ ├── 4.jpg
│ ├── 40.jpg
│ ├── 41.jpg
│ ├── 42.jpg
│ ├── 43.jpg
│ ├── 44.jpg
│ ├── 45.jpg
│ ├── 5.jpg
│ ├── 6.jpg
│ ├── 7.jpg
│ ├── 8.jpg
│ ├── 9.jpg
│ └── AE86.png
├── test.py
├── tools
├── adjust_brightness.py
├── data_loader.py
├── data_mean.py
├── edge_smooth.py
├── get_generator_ckpt.py
├── ops.py
├── utils.py
└── vgg19.py
├── train.py
├── vgg19_weight
└── vgg19.npy file
├── video
├── input
│ └── お花見.mp4
└── output
│ └── お花見.mp4
└── video2anime.py
/.github/FUNDING.yml:
--------------------------------------------------------------------------------
1 | # These are supported funding model platforms
2 |
3 | github: # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
4 | patreon: Asher_Chan
5 | # Replace with a single Patreon username
6 | open_collective: # Replace with a single Open Collective username
7 | ko_fi: # Replace with a single Ko-fi username
8 | tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
9 | community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
10 | liberapay: # Replace with a single Liberapay username
11 | issuehunt: # Replace with a single IssueHunt username
12 | otechie: # Replace with a single Otechie username
13 | lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
14 | custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']
15 |
--------------------------------------------------------------------------------
/AnimeGAN.py:
--------------------------------------------------------------------------------
1 | from tools.ops import *
2 | from tools.utils import *
3 | from glob import glob
4 | import time
5 | import numpy as np
6 | from net import generator
7 | from net.discriminator import D_net
8 | from tools.data_loader import ImageGenerator
9 | from tools.vgg19 import Vgg19
10 |
11 | class AnimeGAN(object) :
12 | def __init__(self, sess, args):
13 | self.model_name = 'AnimeGAN'
14 |
15 | self.sess = sess
16 | self.checkpoint_dir = args.checkpoint_dir
17 | self.log_dir = args.log_dir
18 | self.dataset_name = args.dataset
19 |
20 | self.epoch = args.epoch
21 | self.init_epoch = args.init_epoch # args.epoch // 20
22 |
23 | self.gan_type = args.gan_type
24 |
25 | self.batch_size = args.batch_size
26 |
27 | self.save_freq = args.save_freq
28 |
29 | self.init_lr = args.init_lr
30 | self.d_lr = args.d_lr
31 | self.g_lr = args.g_lr
32 |
33 |
34 | """ Weight """
35 | self.g_adv_weight = args.g_adv_weight
36 | self.d_adv_weight = args.d_adv_weight
37 | self.con_weight = args.con_weight
38 | self.sty_weight = args.sty_weight
39 | self.color_weight = args.color_weight
40 |
41 | self.training_rate = args.training_rate
42 | self.ld = args.ld
43 |
44 | self.img_size = args.img_size
45 | self.img_ch = args.img_ch
46 |
47 | """ Discriminator """
48 | self.n_dis = args.n_dis
49 | self.ch = args.ch
50 | self.sn = args.sn
51 |
52 | self.sample_dir = os.path.join(args.sample_dir, self.model_dir)
53 | check_folder(self.sample_dir)
54 |
55 | self.real = tf.placeholder(tf.float32, [self.batch_size, self.img_size[0], self.img_size[1], self.img_ch], name='real_A')
56 | self.anime = tf.placeholder(tf.float32, [self.batch_size, self.img_size[0], self.img_size[1], self.img_ch], name='anime_A')
57 | self.anime_smooth = tf.placeholder(tf.float32, [self.batch_size, self.img_size[0], self.img_size[1], self.img_ch], name='anime_smooth_A')
58 | self.test_real = tf.placeholder(tf.float32, [1, None, None, self.img_ch], name='test_input')
59 |
60 | self.anime_gray = tf.placeholder(tf.float32, [self.batch_size, self.img_size[0], self.img_size[1], self.img_ch],name='anime_B')
61 |
62 |
63 | self.real_image_generator = ImageGenerator('./dataset/train_photo', self.img_size, self.batch_size)
64 | self.anime_image_generator = ImageGenerator('./dataset/{}'.format(self.dataset_name + '/style'), self.img_size, self.batch_size)
65 | self.anime_smooth_generator = ImageGenerator('./dataset/{}'.format(self.dataset_name + '/smooth'), self.img_size, self.batch_size)
66 | self.dataset_num = max(self.real_image_generator.num_images, self.anime_image_generator.num_images)
67 |
68 | self.vgg = Vgg19()
69 |
70 | print()
71 | print("##### Information #####")
72 | print("# gan type : ", self.gan_type)
73 | print("# dataset : ", self.dataset_name)
74 | print("# max dataset number : ", self.dataset_num)
75 | print("# batch_size : ", self.batch_size)
76 | print("# epoch : ", self.epoch)
77 | print("# init_epoch : ", self.init_epoch)
78 | print("# training image size [H, W] : ", self.img_size)
79 | print("# g_adv_weight,d_adv_weight,con_weight,sty_weight,color_weight : ", self.g_adv_weight,self.d_adv_weight,self.con_weight,self.sty_weight,self.color_weight)
80 | print("# init_lr,g_lr,d_lr : ", self.init_lr,self.g_lr,self.d_lr)
81 | print(f"# training_rate G -- D: {self.training_rate} : 1" )
82 | print()
83 |
84 | ##################################################################################
85 | # Generator
86 | ##################################################################################
87 |
88 | def generator(self,x_init, reuse=False, scope="generator"):
89 |
90 | with tf.variable_scope(scope, reuse=reuse) :
91 | G = generator.G_net(x_init)
92 | return G.fake
93 |
94 | ##################################################################################
95 | # Discriminator
96 | ##################################################################################
97 |
98 | def discriminator(self, x_init, reuse=False, scope="discriminator"):
99 |
100 | D = D_net(x_init, self.ch, self.n_dis, self.sn, reuse=reuse, scope=scope)
101 |
102 | return D
103 |
104 | ##################################################################################
105 | # Model
106 | ##################################################################################
107 | def gradient_panalty(self, real, fake, scope="discriminator"):
108 | if self.gan_type.__contains__('dragan') :
109 | eps = tf.random_uniform(shape=tf.shape(real), minval=0., maxval=1.)
110 | _, x_var = tf.nn.moments(real, axes=[0, 1, 2, 3])
111 | x_std = tf.sqrt(x_var) # magnitude of noise decides the size of local region
112 |
113 | fake = real + 0.5 * x_std * eps
114 |
115 | alpha = tf.random_uniform(shape=[self.batch_size, 1, 1, 1], minval=0., maxval=1.)
116 | interpolated = real + alpha * (fake - real)
117 |
118 | logit, _= self.discriminator(interpolated, reuse=True, scope=scope)
119 |
120 |
121 | grad = tf.gradients(logit, interpolated)[0] # gradient of D(interpolated)
122 | grad_norm = tf.norm(flatten(grad), axis=1) # l2 norm
123 |
124 | GP = 0
125 | # WGAN - LP
126 | if self.gan_type.__contains__('lp'):
127 | GP = self.ld * tf.reduce_mean(tf.square(tf.maximum(0.0, grad_norm - 1.)))
128 |
129 | elif self.gan_type.__contains__('gp') or self.gan_type == 'dragan' :
130 | GP = self.ld * tf.reduce_mean(tf.square(grad_norm - 1.))
131 |
132 | return GP
133 |
134 | def build_model(self):
135 |
136 | """ Define Generator, Discriminator """
137 | self.generated = self.generator(self.real)
138 | self.test_generated = self.generator(self.test_real, reuse=True)
139 |
140 | anime_logit = self.discriminator(self.anime)
141 | anime_gray_logit = self.discriminator(self.anime_gray, reuse=True)
142 |
143 | generated_logit = self.discriminator(self.generated, reuse=True)
144 | smooth_logit = self.discriminator(self.anime_smooth, reuse=True)
145 |
146 | """ Define Loss """
147 | if self.gan_type.__contains__('gp') or self.gan_type.__contains__('lp') or self.gan_type.__contains__('dragan') :
148 | GP = self.gradient_panalty(real=self.anime, fake=self.generated)
149 | else :
150 | GP = 0.0
151 |
152 | # init pharse
153 | init_c_loss = con_loss(self.vgg, self.real, self.generated)
154 | init_loss = self.con_weight * init_c_loss
155 |
156 | self.init_loss = init_loss
157 |
158 | # gan
159 | c_loss, s_loss = con_sty_loss(self.vgg, self.real, self.anime_gray, self.generated)
160 | t_loss = self.con_weight * c_loss + self.sty_weight * s_loss + color_loss(self.real,self.generated) * self.color_weight
161 |
162 | g_loss = self.g_adv_weight * generator_loss(self.gan_type, generated_logit)
163 | d_loss = self.d_adv_weight * discriminator_loss(self.gan_type, anime_logit, anime_gray_logit, generated_logit, smooth_logit) + GP
164 |
165 | self.Generator_loss = t_loss + g_loss
166 | self.Discriminator_loss = d_loss
167 |
168 | """ Training """
169 | t_vars = tf.trainable_variables()
170 | G_vars = [var for var in t_vars if 'generator' in var.name]
171 | D_vars = [var for var in t_vars if 'discriminator' in var.name]
172 |
173 | self.init_optim = tf.train.AdamOptimizer(self.init_lr, beta1=0.5, beta2=0.999).minimize(self.init_loss, var_list=G_vars)
174 | self.G_optim = tf.train.AdamOptimizer(self.g_lr , beta1=0.5, beta2=0.999).minimize(self.Generator_loss, var_list=G_vars)
175 | self.D_optim = tf.train.AdamOptimizer(self.d_lr , beta1=0.5, beta2=0.999).minimize(self.Discriminator_loss, var_list=D_vars)
176 |
177 |
178 | """" Summary """
179 | self.G_loss = tf.summary.scalar("Generator_loss", self.Generator_loss)
180 | self.D_loss = tf.summary.scalar("Discriminator_loss", self.Discriminator_loss)
181 |
182 | self.G_gan = tf.summary.scalar("G_gan", g_loss)
183 | self.G_vgg = tf.summary.scalar("G_vgg", t_loss)
184 | self.G_init_loss = tf.summary.scalar("G_init", init_loss)
185 |
186 | self.V_loss_merge = tf.summary.merge([self.G_init_loss])
187 | self.G_loss_merge = tf.summary.merge([self.G_loss, self.G_gan, self.G_vgg, self.G_init_loss])
188 | self.D_loss_merge = tf.summary.merge([self.D_loss])
189 |
190 |
191 | def train(self):
192 | # initialize all variables
193 | self.sess.run(tf.global_variables_initializer())
194 |
195 | # saver to save model
196 | self.saver = tf.train.Saver(max_to_keep=self.epoch)
197 |
198 | # summary writer
199 | self.writer = tf.summary.FileWriter(self.log_dir + '/' + self.model_dir, self.sess.graph)
200 |
201 | """ Input Image"""
202 | real_img_op, anime_img_op, anime_smooth_op = self.real_image_generator.load_images(), self.anime_image_generator.load_images(), self.anime_smooth_generator.load_images()
203 |
204 |
205 | # restore check-point if it exits
206 | could_load, checkpoint_counter = self.load(self.checkpoint_dir)
207 | if could_load:
208 | start_epoch = checkpoint_counter + 1
209 |
210 | print(" [*] Load SUCCESS")
211 | else:
212 | start_epoch = 0
213 |
214 | print(" [!] Load failed...")
215 |
216 | # loop for epoch
217 | init_mean_loss = []
218 | mean_loss = []
219 | # training times , G : D = self.training_rate : 1
220 | j = self.training_rate
221 | for epoch in range(start_epoch, self.epoch):
222 |
223 | for idx in range(int(self.dataset_num / self.batch_size)):
224 |
225 | anime, anime_smooth, real = self.sess.run([anime_img_op, anime_smooth_op, real_img_op])
226 |
227 | train_feed_dict = {
228 | self.real:real[0],
229 | self.anime:anime[0],
230 | self.anime_gray:anime[1],
231 | self.anime_smooth:anime_smooth[1]
232 | }
233 |
234 | if epoch < self.init_epoch :
235 | # Init G
236 | start_time = time.time()
237 |
238 | real_images, generator_images, _, v_loss, summary_str = self.sess.run([self.real, self.generated,
239 | self.init_optim,
240 | self.init_loss, self.V_loss_merge], feed_dict = train_feed_dict)
241 | self.writer.add_summary(summary_str, epoch)
242 | init_mean_loss.append(v_loss)
243 |
244 | print("Epoch: %3d Step: %5d / %5d time: %f s init_v_loss: %.8f mean_v_loss: %.8f" % (epoch, idx,int(self.dataset_num / self.batch_size), time.time() - start_time, v_loss, np.mean(init_mean_loss)))
245 | if (idx+1)%200 ==0:
246 | init_mean_loss.clear()
247 | else :
248 | start_time = time.time()
249 |
250 | if j == self.training_rate:
251 | # Update D
252 | _, d_loss, summary_str = self.sess.run([self.D_optim, self.Discriminator_loss, self.D_loss_merge],
253 | feed_dict=train_feed_dict)
254 | self.writer.add_summary(summary_str, epoch)
255 |
256 | # Update G
257 | real_images, generator_images, _, g_loss, summary_str = self.sess.run([self.real, self.generated,self.G_optim,
258 | self.Generator_loss, self.G_loss_merge], feed_dict = train_feed_dict)
259 | self.writer.add_summary(summary_str, epoch)
260 |
261 | mean_loss.append([d_loss, g_loss])
262 | if j == self.training_rate:
263 |
264 | print(
265 | "Epoch: %3d Step: %5d / %5d time: %f s d_loss: %.8f, g_loss: %.8f -- mean_d_loss: %.8f, mean_g_loss: %.8f" % (
266 | epoch, idx, int(self.dataset_num / self.batch_size), time.time() - start_time, d_loss, g_loss, np.mean(mean_loss, axis=0)[0],
267 | np.mean(mean_loss, axis=0)[1]))
268 | else:
269 | print(
270 | "Epoch: %3d Step: %5d / %5d time: %f s , g_loss: %.8f -- mean_g_loss: %.8f" % (
271 | epoch, idx, int(self.dataset_num / self.batch_size), time.time() - start_time, g_loss, np.mean(mean_loss, axis=0)[1]))
272 |
273 | if (idx + 1) % 200 == 0:
274 | mean_loss.clear()
275 |
276 | j = j - 1
277 | if j < 1:
278 | j = self.training_rate
279 |
280 |
281 | if (epoch + 1) >= self.init_epoch and np.mod(epoch + 1, self.save_freq) == 0:
282 | self.save(self.checkpoint_dir, epoch)
283 |
284 | if epoch >= self.init_epoch -1:
285 | """ Result Image """
286 | val_files = glob('./dataset/{}/*.*'.format('val'))
287 | save_path = './{}/{:03d}/'.format(self.sample_dir, epoch)
288 | check_folder(save_path)
289 | for i, sample_file in enumerate(val_files):
290 | print('val: '+ str(i) + sample_file)
291 | sample_image = np.asarray(load_test_data(sample_file, self.img_size))
292 | test_real,test_generated = self.sess.run([self.test_real,self.test_generated],feed_dict = {self.test_real:sample_image} )
293 |
294 | save_images(test_real, save_path+'{:03d}_a.jpg'.format(i), None)
295 | save_images(test_generated, save_path + '{:03d}_b.jpg'.format(i), None)
296 | # adjust_brightness_from_photo_to_fake
297 | # save_images(test_generated, save_path+'{:03d}_b.jpg'.format(i), sample_file)
298 |
299 |
300 | @property
301 | def model_dir(self):
302 | return "{}_{}_{}_{}_{}_{}_{}_{}".format(self.model_name, self.dataset_name,
303 | self.gan_type,
304 | int(self.g_adv_weight), int(self.d_adv_weight), int(self.con_weight), int(self.sty_weight), int(self.color_weight))
305 |
306 | def save(self, checkpoint_dir, step):
307 | checkpoint_dir = os.path.join(checkpoint_dir, self.model_dir)
308 |
309 | if not os.path.exists(checkpoint_dir):
310 | os.makedirs(checkpoint_dir)
311 |
312 | self.saver.save(self.sess, os.path.join(checkpoint_dir, self.model_name + '.model'), global_step=step)
313 |
314 | def load(self, checkpoint_dir):
315 | print(" [*] Reading checkpoints...")
316 | checkpoint_dir = os.path.join(checkpoint_dir, self.model_dir)
317 |
318 | ckpt = tf.train.get_checkpoint_state(checkpoint_dir) # checkpoint file information
319 |
320 | if ckpt and ckpt.model_checkpoint_path:
321 | ckpt_name = os.path.basename(ckpt.model_checkpoint_path) # first line
322 | self.saver.restore(self.sess, os.path.join(checkpoint_dir, ckpt_name))
323 | counter = int(ckpt_name.split('-')[-1])
324 | print(" [*] Success to read {}".format(ckpt_name))
325 | return True, counter
326 | else:
327 | print(" [*] Failed to find a checkpoint")
328 | return False, 0
329 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # AnimeGAN
2 | A Tensorflow implementation of AnimeGAN for fast photo animation !     [日本語](https://github.com/TachibanaYoshino/AnimeGAN/blob/master/doc/Japanese_README.md)
3 | The paper can be accessed [here](https://github.com/TachibanaYoshino/AnimeGAN/blob/master/doc/Chen2020_Chapter_AnimeGAN.pdf) or on the [website](https://link.springer.com/chapter/10.1007/978-981-15-5577-0_18).
4 |
5 | ### If you like what I'm doing you can tip me on [*patreon*](https://www.patreon.com/Asher_Chan).
6 |
7 | Photos [](https://drive.google.com/file/d/1PbBkmj1EhULvEE8AXr2z84pZ2DQJN4hc/view?usp=sharing)
8 |
9 | Videos [](https://drive.google.com/file/d/1qhBxA72Wxbh6Eyhd-V0zY_jTIblP9rHz/view?usp=sharing)
10 | ____
11 | * Added the AnimeGAN [Colab](https://drive.google.com/file/d/1PbBkmj1EhULvEE8AXr2z84pZ2DQJN4hc/view?usp=sharing).
12 | * [**AnimeGANv3**](https://github.com/TachibanaYoshino/AnimeGANv3) has been released.
13 | * [**AnimeGANv2, the improved version of AnimeGAN.**](https://github.com/TachibanaYoshino/AnimeGANv2)
14 | * **Online access**: Be grateful to [@TonyLianLong](https://github.com/TonyLianLong/AnimeGAN.js) for developing an online access project, you can implement photo animation through a browser without installing anything, [click here to have a try](https://animegan.js.org/).
15 | * **Pytorch version**: [pytorch-animeGAN](https://github.com/ptran1203/pytorch-animeGAN). Be grateful to @ptran1203.
16 |
17 | -----
18 | **Some suggestions:**
19 | 1. since the real photos in the training set are all landscape photos, if you want to stylize the photos with people as the main body, you may as well add at least 3000 photos of people in the training set and retrain to obtain a new model.
20 | 2. In order to obtain a better face animation effect, when using 2 images as data pairs for training, it is suggested that the faces in the photos and the faces in the anime style data should be consistent in terms of gender as much as possible.
21 | 3. The generated stylized images will be affected by the overall brightness and tone of the style data, so try not to select the anime images of night as the style data, and it is necessary to make an exposure compensation for the overall style data to promote the consistency of brightness and darkness of the entire style data.
22 |
23 | **News:**
24 |       ***AnimeGANv2*** has been released and can be accessed [here](https://github.com/TachibanaYoshino/AnimeGANv2).
25 | ```yaml
26 | The improvement directions of AnimeGANv2 mainly include the following 4 points:
27 | ```
28 |    1. Solve the problem of high-frequency artifacts in the generated image.
29 |    2. It is easy to train and directly achieve the effects in the paper.
30 |    3. Further reduce the number of parameters of the generator network.
31 |    4. Use new high-quality style data, which come from BD movies as much as possible.
32 |
33 | ___
34 |
35 | ## Requirements
36 | - python 3.7
37 | - tensorflow-gpu 1.15.0 (ubuntu, GPU 2080Ti, cuda 10.0.130, cudnn 7.6.0)
38 | - opencv
39 | - tqdm
40 | - numpy
41 | - glob
42 | - argparse
43 |
44 | ## Usage
45 | ### 1. Inference
46 | e.g. `python test.py --checkpoint_dir checkpoint/generator_Hayao_weight --test_dir dataset/test/real --style_name H`
47 |
48 | ### 2. Convert video to anime
49 | e.g. `python video2anime.py --video video/input/お花見.mp4 --checkpoint_dir ./checkpoint/generator_Hayao_weight`
50 |
51 | ### 3. Train
52 | #### 1. Download vgg19 or Pretrained model
53 | > [vgg19.npy](https://github.com/TachibanaYoshino/AnimeGAN/releases/tag/vgg16%2F19.npy)
54 |
55 | > [Pretrained model](https://github.com/TachibanaYoshino/AnimeGAN/releases/tag/Haoyao-style_V1.0)
56 |
57 | #### 2. Download dataset
58 | > [Link](https://github.com/TachibanaYoshino/AnimeGAN/releases/tag/dataset-1)
59 |
60 | #### 3. Do edge_smooth
61 | e.g. `python edge_smooth.py --dataset Hayao --img_size 256`
62 |
63 | #### 4. Train
64 | e.g. `python train.py --dataset Hayao --epoch 101 --init_epoch 5`
65 |
66 | #### 5. Extract the weights of the generator
67 | e.g. `python get_generator_ckpt.py --checkpoint_dir ../checkpoint/AnimeGAN_Hayao_lsgan_300_300_1_1_10 --style_name Hayao`
68 |
69 |
70 | ____
71 | ## Results
72 | :blush: pictures from the paper - *AnimeGAN: a novel lightweight GAN for photo animation*
73 |
74 | 
75 | 
76 | 
77 | 
78 |
79 | :heart_eyes: Photo to Hayao Style
80 |
81 |  
82 |  
83 |  
84 |  
85 |  
86 |  
87 |  
88 |  
89 |  
90 |  
91 |  
92 | ____
93 | ## License
94 | This repo is made freely available to academic and non-academic entities for non-commercial purposes such as academic research, teaching, scientific publications. Permission is granted to use the AnimeGAN given that you agree to my license terms. Regarding the request for commercial use, please contact us via email to help you obtain the authorization letter.
95 | ## Author
96 | Xin Chen, Gang Liu, Jie Chen
97 | ## Acknowledgment
98 | This code is based on the [CartoonGAN-Tensorflow](https://github.com/taki0112/CartoonGAN-Tensorflow/blob/master/CartoonGAN.py) and [Anime-Sketch-Coloring-with-Swish-Gated-Residual-UNet](https://github.com/pradeeplam/Anime-Sketch-Coloring-with-Swish-Gated-Residual-UNet). Thanks to the contributors of this project.
99 |
100 |
--------------------------------------------------------------------------------
/Requirements installation tutorial.md:
--------------------------------------------------------------------------------
1 | 1. Linux os (Ubuntu 18.04 or 20.04), Nvidia GPU (e.g. 2080ti) , Install GPU Driver
2 | 2. Use Anaconda or Miniconda to build python environment. The Miniconda version I used can be downloaded [here](https://repo.anaconda.com/miniconda/Miniconda3-py37_4.9.2-Linux-x86_64.sh).
3 | 3. Install CUDA in conda command window : `conda install cudatoolkit=10` , Then install cudnn : `conda install cudnn=7.6.0`
4 | 4. Install other requirements by pip command, such as `pip install tensorflow-gpu==1.15.0`, `pip install tqdm`, `pip install glob` and so on.
5 | 5. Place the downloaded dataset in the corresponding directory, and then perform training and testing.
6 |
7 |
--------------------------------------------------------------------------------
/checkpoint/AnimeGAN_Hayao_lsgan_300_300_1_2_10/checkpoints files:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/checkpoint/generator_Hayao_weight/Hayao-60.ckpt.data-00000-of-00001:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/checkpoint/generator_Hayao_weight/Hayao-60.ckpt.data-00000-of-00001
--------------------------------------------------------------------------------
/checkpoint/generator_Hayao_weight/Hayao-60.ckpt.index:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/checkpoint/generator_Hayao_weight/Hayao-60.ckpt.index
--------------------------------------------------------------------------------
/checkpoint/generator_Hayao_weight/Hayao-60.ckpt.meta:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/checkpoint/generator_Hayao_weight/Hayao-60.ckpt.meta
--------------------------------------------------------------------------------
/checkpoint/generator_Hayao_weight/checkpoint:
--------------------------------------------------------------------------------
1 | model_checkpoint_path: "Hayao-60.ckpt"
2 | all_model_checkpoint_paths: "Hayao-60.ckpt"
3 |
--------------------------------------------------------------------------------
/dataset/Hayao/smooth/smooth dir:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/Hayao/style/style images dir:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/Paprika/smooth dir/smooth images:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/Paprika/style/style images dir:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/Shinkai/sytle/sytle images dir:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/test/HR_photo/HR photos:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/test/real/photos:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/test/test_photo/none:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/test/test_photo256/256x256 photos:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/train_photo/train photos:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/dataset/val/valid images:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/doc/3-1block.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/3-1block.png
--------------------------------------------------------------------------------
/doc/Chen2020_Chapter_AnimeGAN.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/Chen2020_Chapter_AnimeGAN.pdf
--------------------------------------------------------------------------------
/doc/GD.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/GD.png
--------------------------------------------------------------------------------
/doc/Japanese_README.md:
--------------------------------------------------------------------------------
1 | # AnimeGAN
2 | このAnimeGANは写真をアニメに素早く変換するために使用できます!
3 | 論文は[ここ](https://github.com/TachibanaYoshino/AnimeGAN/blob/master/doc/Chen2020_Chapter_AnimeGAN.pdf) あるいはこの[サイト](https://link.springer.com/chapter/10.1007/978-981-15-5577-0_18) からアクセスできます。
4 |
5 | [**AnimeGANの改良版である AnimeGANv2.**](https://github.com/TachibanaYoshino/AnimeGANv2)
6 | **オンラインアクセス**： このオンラインサイトを開発してくれた[@TonyLianLong](https://github.com/TonyLianLong/AnimeGAN.js) のおかげで、何もインストールしなくてもブラウザから直接アクセスできます。[ここをクリックして試してみてください！](https://animegan.js.org/)
7 | **Pytorch version**: [pytorch-animeGAN](https://github.com/ptran1203/pytorch-animeGAN).
8 |
9 | -----
10 | **いくつかのアドバイス**
11 | 1. 訓練セットの真実の写真は風景写真ですので、人物を主体として写真を様式化するなら、トレーニングセットに少なくとも3000人の写真を追加して、新たなモデルを獲得するようにトレーニングしてください。
12 | 2. より良い顔アニメーション効果を得るために、2枚の画像をデータとして使ってトレーニングする場合、写真の顔とアニメスタイルデータの中の顔は性別においてできるだけ一致するようにすることをおすすめします。
13 | 3. 生成された画風化画像は、スタイルデータの全体的な輝度と色調の影響を受けるので、できるだけ夜のアニメ画像をスタイルデータとして選択しないでください。また、全体のスタイルデータに対して露出補償を行う必要があります。全体のスタイルデータの明暗一致を促進します。
14 |
15 | **新しいニュース**：
16 |       **AnimeGANv2** がリリースされ、[**ここ**](https://github.com/TachibanaYoshino/AnimeGANv2)からアクセスできます。
17 | ```yaml
18 | AnimeGANv2 の改善方針には、主に以下の4点があります。
19 | ```
20 |    1. 生成された画像の高周波アーティファクトの問題を解決する。
21 |    2. トレーニングが簡単で、論文の効果を直接達成できます。
22 |    3. ジェネレーターネットワークのパラメーターの数をさらに減らします。
23 |    4. 可能な限りBDムービーからの新しい高品質スタイルデータを使用します。
24 |
25 | ___
26 |
27 | ## インストール依存項
28 | - python 3.7
29 | - tensorflow-gpu 1.15.0 (ubuntu, GPU 2080Ti, cuda 10.0.130, cudnn 7.6.0)
30 | - opencv
31 | - tqdm
32 | - numpy
33 | - glob
34 | - argparse
35 |
36 | ## 使い方
37 | ### 1. テスト
38 | e.g. `python test.py --checkpoint_dir checkpoint/generator_Hayao_weight --test_dir dataset/test/real --style_name H`
39 |
40 | ### 2. ビデオをアニメに変換します
41 | e.g. `python video2anime.py --video video/input/お花見.mp4 --checkpoint_dir ./checkpoint/generator_Hayao_weight`
42 |
43 | ### 3.トレーニング
44 | #### 1. vgg 19あるいは予備トレーニングのモデルをダウンロードします
45 | > [vgg19.npy](https://github.com/TachibanaYoshino/AnimeGAN/releases/tag/vgg16%2F19.npy)
46 |
47 | > [Pretrained model](https://github.com/TachibanaYoshino/AnimeGAN/releases/tag/Haoyao-style_v1.0)
48 |
49 | #### 2. データセットをダウンロード
50 | > [Link](https://github.com/TachibanaYoshino/AnimeGAN/releases/tag/dataset-1)
51 |
52 | #### 3. エッジブラーを実行する
53 | e.g. `python edge_smooth.py --dataset Hayao --img_size 256`
54 |
55 | #### 4. トレーニング
56 | e.g. `python train.py --dataset Hayao --epoch 101 --init_epoch 5`
57 |
58 | #### 5. ジェネレーターのパラメーターを抽出する
59 | e.g. `python get_generator_ckpt.py --checkpoint_dir ../checkpoint/AnimeGAN_Hayao_lsgan_300_300_1_1_10 --style_name Hayao`
60 |
61 | ____
62 | ## 結果
63 | :blush: 論文からの画像。
64 |
65 | 
66 | 
67 | 
68 | 
69 |
70 | :heart_eyes: 写真は宮崎さんのアニメスタイルになります。
71 |
72 |  
73 |  
74 |  
75 |  
76 |  
77 |  
78 |  
79 |  
80 |  
81 |  
82 |  
83 | _____
84 | ## ライセンス
85 | このレポは、学術研究、教育、科学出版物などの非営利目的で、学術団体および非学術団体が自由に利用できるようになっています。
86 | 私のライセンス条項に同意すると、AnimeGANの使用が許可されます。 商用利用のリクエストについては、承認書を入手するためにメールでご連絡ください。
87 | ## 著者
88 | Xin Chen, Gang Liu, Jie Chen
89 | ## 感謝
90 | このコードは[CartoonGAN-Tensorflow](https://github.com/taki0112/CartoonGAN-Tensorflow/blob/master/CartoonGAN.py) と [Anime-Sketch-Coloring-with-Swish-Gated-Residual-UNet](https://github.com/pradeeplam/Anime-Sketch-Coloring-with-Swish-Gated-Residual-UNet) に基づいています。彼らの貢献に感謝します。
91 |
92 |
--------------------------------------------------------------------------------
/doc/e2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/e2.png
--------------------------------------------------------------------------------
/doc/e3.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/e3.png
--------------------------------------------------------------------------------
/doc/e4.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/e4.png
--------------------------------------------------------------------------------
/doc/sota.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/doc/sota.png
--------------------------------------------------------------------------------
/net/discriminator.py:
--------------------------------------------------------------------------------
1 |
2 | from tools.ops import *
3 |
4 | def D_net(x_init,ch, n_dis,sn, scope, reuse):
5 | channel = ch // 2
6 | with tf.variable_scope(scope, reuse=reuse):
7 | x = conv(x_init, channel, kernel=3, stride=1, pad=1, use_bias=False, sn=sn, scope='conv_0')
8 | x = lrelu(x, 0.2)
9 |
10 | for i in range(1, n_dis):
11 | x = conv(x, channel * 2, kernel=3, stride=2, pad=1, use_bias=False, sn=sn, scope='conv_s2_' + str(i))
12 | x = lrelu(x, 0.2)
13 |
14 | x = conv(x, channel * 4, kernel=3, stride=1, pad=1, use_bias=False, sn=sn, scope='conv_s1_' + str(i))
15 | x = instance_norm(x, scope='ins_norm_' + str(i))
16 | x = lrelu(x, 0.2)
17 |
18 | channel = channel * 2
19 |
20 | x = conv(x, channel * 2, kernel=3, stride=1, pad=1, use_bias=False, sn=sn, scope='last_conv')
21 | x = instance_norm(x, scope='last_ins_norm')
22 | x = lrelu(x, 0.2)
23 |
24 | x = conv(x, channels=1, kernel=3, stride=1, pad=1, use_bias=False, sn=sn, scope='D_logit')
25 |
26 | return x
27 |
28 |
--------------------------------------------------------------------------------
/net/generator.py:
--------------------------------------------------------------------------------
1 | from tools.ops import *
2 | import tensorflow as tf
3 |
4 |
5 | def Conv2D(inputs, filters, kernel_size=3, strides=1, padding='VALID', Use_bias = None):
6 | if kernel_size == 3:
7 | inputs = tf.pad(inputs, [[0, 0], [1, 1], [1, 1], [0, 0]], mode="REFLECT")
8 | return tf.contrib.layers.conv2d(
9 | inputs,
10 | num_outputs=filters,
11 | kernel_size=kernel_size,
12 | stride=strides,
13 | biases_initializer= Use_bias,
14 | normalizer_fn=None,
15 | activation_fn=None,
16 | padding=padding)
17 |
18 |
19 | def Conv2DNormLReLU(inputs, filters, kernel_size=3, strides=1, padding='VALID', Use_bias = None):
20 | x = Conv2D(inputs, filters, kernel_size, strides,padding=padding, Use_bias = Use_bias)
21 | x = instance_norm(x,scope=None)
22 | return lrelu(x)
23 |
24 | def dwise_conv(input, k_h=3, k_w=3, channel_multiplier=1, strides=[1, 1, 1, 1],
25 | padding='VALID', stddev=0.02, name='dwise_conv', bias=False):
26 | input = tf.pad(input, [[0, 0], [1, 1], [1, 1], [0, 0]], mode="REFLECT")
27 | with tf.variable_scope(name):
28 | in_channel = input.get_shape().as_list()[-1]
29 | w = tf.get_variable('w', [k_h, k_w, in_channel, channel_multiplier],regularizer=None,initializer=tf.truncated_normal_initializer(stddev=stddev))
30 | conv = tf.nn.depthwise_conv2d(input, w, strides, padding, rate=None, name=name, data_format=None)
31 | if bias:
32 | biases = tf.get_variable('bias', [in_channel * channel_multiplier],initializer=tf.constant_initializer(0.0))
33 | conv = tf.nn.bias_add(conv, biases)
34 | return conv
35 |
36 | def Separable_conv2d(inputs, filters, kernel_size=3, strides=1, padding='VALID', Use_bias = None):
37 | if kernel_size==3 and strides==1:
38 | inputs = tf.pad(inputs, [[0, 0], [1, 1], [1, 1], [0, 0]], mode="REFLECT")
39 | if strides == 2:
40 | inputs = tf.pad(inputs, [[0, 0], [0, 1], [0, 1], [0, 0]], mode="REFLECT")
41 | return tf.contrib.layers.separable_conv2d(
42 | inputs,
43 | num_outputs=filters,
44 | kernel_size=kernel_size,
45 | depth_multiplier=1,
46 | stride=strides,
47 | biases_initializer=Use_bias,
48 | normalizer_fn=tf.contrib.layers.instance_norm,
49 | activation_fn=lrelu,
50 | padding=padding)
51 |
52 | def Conv2DTransposeLReLU(inputs, filters, kernel_size=2, strides=2, padding='SAME', Use_bias = None):
53 |
54 | return tf.contrib.layers.conv2d_transpose(inputs,
55 | num_outputs=filters,
56 | kernel_size=kernel_size,
57 | stride=strides,
58 | biases_initializer=Use_bias,
59 | normalizer_fn=tf.contrib.layers.instance_norm,
60 | activation_fn=lrelu,
61 | padding=padding)
62 |
63 | def Unsample(inputs, filters, kernel_size=3):
64 | '''
65 | An alternative to transposed convolution where we first resize, then convolve.
66 | See http://distill.pub/2016/deconv-checkerboard/
67 | For some reason the shape needs to be statically known for gradient propagation
68 | through tf.image.resize_images, but we only know that for fixed image size, so we
69 | plumb through a "training" argument
70 | '''
71 | new_H, new_W = 2 * tf.shape(inputs)[1], 2 * tf.shape(inputs)[2]
72 | inputs = tf.image.resize_images(inputs, [new_H, new_W])
73 |
74 | return Separable_conv2d(filters=filters, kernel_size=kernel_size, inputs=inputs)
75 |
76 | def Downsample(inputs, filters = 256, kernel_size=3):
77 | '''
78 | An alternative to transposed convolution where we first resize, then convolve.
79 | See http://distill.pub/2016/deconv-checkerboard/
80 | For some reason the shape needs to be statically known for gradient propagation
81 | through tf.image.resize_images, but we only know that for fixed image size, so we
82 | plumb through a "training" argument
83 | '''
84 |
85 | new_H, new_W = tf.shape(inputs)[1] // 2, tf.shape(inputs)[2] // 2
86 | inputs = tf.image.resize_images(inputs, [new_H, new_W])
87 |
88 | return Separable_conv2d(filters=filters, kernel_size=kernel_size, inputs=inputs)
89 |
90 |
91 |
92 |
93 | class G_net(object):
94 |
95 | def __init__(self, inputs):
96 |
97 | with tf.variable_scope('G_MODEL'):
98 |
99 | with tf.variable_scope('b1'):
100 | inputs = Conv2DNormLReLU(inputs, 64)
101 | inputs = Conv2DNormLReLU(inputs, 64)
102 | inputs = Separable_conv2d(inputs,128,strides=2) + Downsample(inputs, 128)
103 |
104 | with tf.variable_scope('b2'):
105 | inputs = Conv2DNormLReLU(inputs, 128)
106 | inputs = Separable_conv2d(inputs, 128)
107 | inputs = Separable_conv2d(inputs, 256, strides=2) + Downsample(inputs, 256)
108 |
109 | with tf.variable_scope('m'):
110 | inputs = Conv2DNormLReLU(inputs, 256)
111 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r1')
112 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r2')
113 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r3')
114 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r4')
115 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r5')
116 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r6')
117 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r7')
118 | inputs = self.InvertedRes_block(inputs, 2, 256, 1, 'r8')
119 | inputs = Conv2DNormLReLU(inputs, 256)
120 | with tf.variable_scope('u2'):
121 | inputs = Unsample(inputs, 128)
122 | inputs = Separable_conv2d(inputs, 128)
123 | inputs = Conv2DNormLReLU(inputs, 128)
124 |
125 | with tf.variable_scope('u1'):
126 | inputs = Unsample(inputs,128) # The number of the filters in this layer is 128 while it is 64 in the graph of the paper. Please refer to the code.
127 | inputs = Conv2DNormLReLU(inputs, 64)
128 | inputs = Conv2DNormLReLU(inputs, 64)
129 |
130 | out = Conv2D(inputs, filters =3, kernel_size=1, strides=1)
131 | self.fake = tf.tanh(out)
132 |
133 |
134 | def InvertedRes_block(self, input, expansion_ratio, output_dim, stride, name, reuse=False, bias=None):
135 | with tf.variable_scope(name, reuse=reuse):
136 | # pw
137 | bottleneck_dim = round(expansion_ratio * input.get_shape().as_list()[-1])
138 | net = Conv2DNormLReLU(input, bottleneck_dim, kernel_size=1, Use_bias=bias)
139 |
140 | # dw
141 | net = dwise_conv(net, name=name)
142 | net = instance_norm(net,scope='1')
143 | net = lrelu(net)
144 |
145 | # pw & linear
146 | net = Conv2D(net, output_dim, kernel_size=1)
147 | net = instance_norm(net,scope='2')
148 |
149 | # element wise add, only for stride==1
150 | if (int(input.get_shape().as_list()[-1]) == output_dim) and stride == 1:
151 | net = input + net
152 |
153 | return net
154 |
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/105.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/105.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/185.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/185.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/19.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/19.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/190.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/190.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/198.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/198.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/204.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/204.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/208.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/208.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/246.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/246.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/256.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/256.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/258.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/258.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/26.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/26.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/35.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/35.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/447.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/447.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/539.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/539.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/599.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/599.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/618.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/618.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/632.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/632.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/675.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/675.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/681.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/681.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/69.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/69.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/75.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/75.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/AnimeGAN/762.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/AnimeGAN/762.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/105_1562736508.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/105_1562736508.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/185_1562736844.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/185_1562736844.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/190_1562737136.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/190_1562737136.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/198_1562737236.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/198_1562737236.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/19_1562736964.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/19_1562736964.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/204_1562737290.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/204_1562737290.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/208_1562737392.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/208_1562737392.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/246_1562737460.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/246_1562737460.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/256_1562739338.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/256_1562739338.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/258_1562737714.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/258_1562737714.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/26_1562737803.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/26_1562737803.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/35_1562737865.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/35_1562737865.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/447_1562737943.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/447_1562737943.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/539_1562738321.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/539_1562738321.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/599_1562739505.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/599_1562739505.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/618_1562739199.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/618_1562739199.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/632_1562738596.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/632_1562738596.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/675_1562738662.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/675_1562738662.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/681_1562738765.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/681_1562738765.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/69_1562738956.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/69_1562738956.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/75_1562739149.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/75_1562739149.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/CartoonGAN/762_1562739095.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/CartoonGAN/762_1562739095.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/105.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/105.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/185.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/185.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/19.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/19.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/190.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/190.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/198.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/198.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/204.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/204.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/208.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/208.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/246.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/246.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/256.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/256.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/258.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/258.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/26.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/26.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/35.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/35.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/447.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/447.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/539.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/539.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/599.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/599.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/618.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/618.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/632.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/632.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/675.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/675.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/681.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/681.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/69.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/69.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/75.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/75.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/ComixGAN/762.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/ComixGAN/762.png
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/105.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/105.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/185.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/185.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/19.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/19.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/190.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/190.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/198.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/198.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/204.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/204.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/208.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/208.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/246.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/246.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/256.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/256.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/258.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/258.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/26.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/26.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/35.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/35.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/447.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/447.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/539.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/539.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/599.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/599.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/618.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/618.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/632.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/632.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/675.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/675.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/681.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/681.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/69.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/69.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/75.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/75.jpg
--------------------------------------------------------------------------------
/result/Hayao/SOTA_comparison/real/762.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/SOTA_comparison/real/762.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 1.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 10.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 10.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 11.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 11.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 12.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 12.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 13.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 13.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 14.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 14.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 15.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 15.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 16.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 16.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 17.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 17.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 18.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 18.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 19.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 19.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 2.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 2.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 20.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 20.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 21.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 21.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 22.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 22.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 23.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 23.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 24.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 24.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 25.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 25.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 26.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 26.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 27.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 27.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 28.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 28.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 29.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 29.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 3.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 3.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 30.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 30.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 31.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 31.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 32.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 32.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 33.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 33.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 34.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 34.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 35.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 35.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 36.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 36.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 37.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 37.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 38.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 38.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 39.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 39.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 4.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 4.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 40.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 40.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 41.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 41.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 42.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 42.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 43.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 43.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 44.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 44.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 45.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 45.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 5.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 5.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 6.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 6.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 7.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 7.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 8.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 8.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/ 9.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/ 9.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo/AE86.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo/AE86.png
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 1.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 10.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 10.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 11.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 11.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 12.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 12.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 13.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 13.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 14.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 14.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 15.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 15.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 16.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 16.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 17.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 17.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 18.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 18.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 19.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 19.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 2.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 2.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 20.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 20.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 21.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 21.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 22.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 22.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 23.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 23.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 24.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 24.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 25.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 25.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 26.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 26.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 27.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 27.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 28.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 28.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 29.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 29.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 3.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 3.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 30.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 30.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 31.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 31.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 32.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 32.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 33.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 33.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 34.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 34.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 35.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 35.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 36.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 36.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 37.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 37.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 38.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 38.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 39.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 39.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 4.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 4.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 40.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 40.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 41.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 41.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 42.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 42.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 43.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 43.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 44.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 44.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 45.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 45.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 5.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 5.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 6.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 6.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 7.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 7.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 8.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 8.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/ 9.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/ 9.jpg
--------------------------------------------------------------------------------
/result/Hayao/photo_result/AE86.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/result/Hayao/photo_result/AE86.png
--------------------------------------------------------------------------------
/test.py:
--------------------------------------------------------------------------------
1 | import argparse
2 | from tools.utils import *
3 | import os
4 | from tqdm import tqdm
5 | from glob import glob
6 | import time
7 | import numpy as np
8 | from net import generator
9 | os.environ["CUDA_VISIBLE_DEVICES"] = "0"
10 |
11 | def parse_args():
12 | desc = "AnimeGAN"
13 | parser = argparse.ArgumentParser(description=desc)
14 |
15 | parser.add_argument('--checkpoint_dir', type=str, default='checkpoint/'+'generator_Hayao_weight/',
16 | help='Directory name to save the checkpoints')
17 | parser.add_argument('--test_dir', type=str, default='dataset/test/HR_photo',
18 | help='Directory name of test photos')
19 | parser.add_argument('--style_name', type=str, default='Hayao',
20 | help='what style you want to get')
21 | parser.add_argument('--if_adjust_brightness', type=bool, default=False,
22 | help='adjust brightness by the real photo')
23 |
24 | return parser.parse_args()
25 |
26 | def stats_graph(graph):
27 | flops = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.float_operation())
28 | # params = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.trainable_variables_parameter())
29 | print('FLOPs: {}'.format(flops.total_float_ops))
30 |
31 | def test(checkpoint_dir, style_name, test_dir, if_adjust_brightness, img_size=[256,256]):
32 | # tf.reset_default_graph()
33 | test_files = glob('{}/*.*'.format(test_dir))
34 |
35 | result_dir = 'results/'+style_name
36 | check_folder(result_dir)
37 |
38 | # test_real = tf.placeholder(tf.float32, [1, 256, 256, 3], name='test')
39 | test_real = tf.placeholder(tf.float32, [1, None, None, 3], name='test')
40 |
41 | with tf.variable_scope("generator", reuse=False):
42 | test_generated = generator.G_net(test_real).fake
43 |
44 | generator_var = [var for var in tf.trainable_variables() if var.name.startswith('generator')]
45 | saver = tf.train.Saver(generator_var)
46 |
47 | gpu_options = tf.GPUOptions(allow_growth=True)
48 | with tf.Session(config=tf.ConfigProto(allow_soft_placement=True, gpu_options=gpu_options)) as sess:
49 | # tf.global_variables_initializer().run()
50 | # load model
51 | ckpt = tf.train.get_checkpoint_state(checkpoint_dir) # checkpoint file information
52 | if ckpt and ckpt.model_checkpoint_path:
53 | ckpt_name = os.path.basename(ckpt.model_checkpoint_path) # first line
54 | saver.restore(sess, os.path.join(checkpoint_dir, ckpt_name))
55 | print(" [*] Success to read {}".format(ckpt_name))
56 | else:
57 | print(" [*] Failed to find a checkpoint")
58 | return
59 |
60 | # FLOPs
61 | # stats_graph(tf.get_default_graph())
62 |
63 | begin = time.time()
64 | for sample_file in tqdm(test_files) :
65 | # print('Processing image: ' + sample_file)
66 | sample_image = np.asarray(load_test_data(sample_file, img_size))
67 | image_path = os.path.join(result_dir,'{0}'.format(os.path.basename(sample_file)))
68 | fake_img = sess.run(test_generated, feed_dict = {test_real : sample_image})
69 | if if_adjust_brightness:
70 | save_images(fake_img, image_path, sample_file)
71 | else:
72 | save_images(fake_img, image_path, None)
73 | end = time.time()
74 | print(f'test-time: {end-begin} s')
75 | print(f'one image test time : {(end-begin)/len(test_files)} s')
76 | print(f'result path: {result_dir}')
77 | if __name__ == '__main__':
78 | arg = parse_args()
79 | print(arg.checkpoint_dir)
80 | test(arg.checkpoint_dir, arg.style_name, arg.test_dir, arg.if_adjust_brightness)
81 |
--------------------------------------------------------------------------------
/tools/adjust_brightness.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 | import cv2
3 |
4 | def read_img(image_path):
5 | img = cv2.imread(image_path)
6 | img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
7 | assert len(img.shape)==3
8 | return img
9 |
10 |
11 | # Calculates the average brightness in the specified irregular image
12 | def calculate_average_brightness(img):
13 | # Average value of three color channels
14 | B = img[..., 0].mean()
15 | G = img[..., 1].mean()
16 | R = img[..., 2].mean()
17 |
18 | brightness = 0.299 * R + 0.587 * G + 0.114 * B
19 | return brightness, B, G, R
20 |
21 | # Adjusting the average brightness of the target image to the average brightness of the source image
22 | def adjust_brightness_from_src_to_dst(dst, src,path=None,if_show=None, if_info=None):
23 | brightness1, B1, G1, R1 = calculate_average_brightness(src)
24 | brightness2, B2, G2, R2 = calculate_average_brightness(dst)
25 | brightness_difference = brightness1 / brightness2
26 |
27 | if if_info:
28 | print('Average brightness of original image', brightness1)
29 | print('Average brightness of target', brightness2)
30 | print('Brightness Difference between Original Image and Target', brightness_difference)
31 |
32 | # According to the average display brightness
33 | dstf = dst * brightness_difference
34 |
35 | # According to the average value of the three-color channel
36 | # dstf = dst.copy().astype(np.float32)
37 | # dstf[..., 0] = dst[..., 0] * (B1 / B2)
38 | # dstf[..., 1] = dst[..., 1] * (G1 / G2)
39 | # dstf[..., 2] = dst[..., 2] * (R1 / R2)
40 |
41 | # To limit the results and prevent crossing the border,
42 | # it must be converted to uint8, otherwise the default result is float32, and errors will occur.
43 | dstf = np.clip(dstf, 0, 255)
44 | dstf = np.uint8(dstf)
45 |
46 | ma,na,_ = src.shape
47 | mb,nb,_ = dst.shape
48 | result_show_img = np.zeros((max(ma, mb), 3 * max(na, nb), 3))
49 | result_show_img[:mb, :nb, :] = dst
50 | result_show_img[:ma, nb:nb + na, :] = src
51 | result_show_img[:mb, nb + na:nb + na + nb, :] = dstf
52 | result_show_img = result_show_img.astype(np.uint8)
53 |
54 | if if_show:
55 | cv2.imshow('-', cv2.cvtColor(result_show_img, cv2.COLOR_BGR2RGB))
56 | cv2.waitKey(0)
57 | cv2.destroyAllWindows()
58 |
59 | if path != None:
60 | cv2.imwrite(path, cv2.cvtColor(result_show_img, cv2.COLOR_BGR2RGB))
61 |
62 | return dstf
63 |
64 | if __name__ == '__main__':
65 |
66 | A = read_img('../Brightness_tool/A.png')
67 | B = read_img('../Brightness_tool/B.png')
68 |
69 | # adjusted = adjust_brightness_from_src_to_dst(A, B, path='resA.png')
70 | adjusted = adjust_brightness_from_src_to_dst(A, B)
71 |
--------------------------------------------------------------------------------
/tools/data_loader.py:
--------------------------------------------------------------------------------
1 | import os
2 | import tensorflow as tf
3 | import cv2,random
4 | import numpy as np
5 |
6 |
7 | class ImageGenerator(object):
8 |
9 | def __init__(self, image_dir,size, batch_size, num_cpus = 16):
10 | self.paths = self.get_image_paths_train(image_dir)
11 | self.num_images = len(self.paths)
12 | self.num_cpus = num_cpus
13 | self.size = size
14 | self.batch_size = batch_size
15 |
16 | def get_image_paths_train(self, image_dir):
17 | paths = []
18 | for path in os.listdir(image_dir):
19 | # Check extensions of filename
20 | if path.split('.')[-1] not in ['jpg', 'jpeg', 'png', 'gif']:
21 | continue
22 |
23 | # Construct complete path to anime image
24 | path_full = os.path.join(image_dir, path)
25 |
26 | # Validate if colorized image exists
27 | if not os.path.isfile(path_full):
28 | continue
29 | paths.append(path_full)
30 | return paths
31 |
32 | def read_image(self, img_path1):
33 |
34 | if 'style' in img_path1.decode() or 'smooth' in img_path1.decode():
35 | # color image1
36 | image1 = cv2.imread(img_path1.decode()).astype(np.float32)
37 | image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
38 |
39 | # gray image2
40 | image2 = cv2.imread(img_path1.decode(),cv2.IMREAD_GRAYSCALE).astype(np.float32)
41 | image2 = np.asarray([image2,image2,image2])
42 | image2= np.transpose(image2,(1,2,0))
43 |
44 | else:
45 | # color image1
46 | image1 = cv2.imread(img_path1.decode()).astype(np.float32)
47 | image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
48 |
49 | image2 = np.zeros(image1.shape).astype(np.float32)
50 |
51 | return image1, image2
52 |
53 | def load_image(self, img1 ):
54 | image1, image2 = self.read_image(img1)
55 | processing_image1 = image1/ 127.5 - 1.0
56 | processing_image2 = image2/ 127.5 - 1.0
57 | return (processing_image1,processing_image2)
58 |
59 | def load_images(self):
60 |
61 | dataset = tf.data.Dataset.from_tensor_slices(self.paths)
62 |
63 | # Repeat indefinitely
64 | dataset = dataset.repeat()
65 |
66 | # Unform shuffle
67 | dataset = dataset.shuffle(buffer_size=len(self.paths))
68 |
69 | # Map path to image
70 | dataset = dataset.map(lambda img: tf.py_func(
71 | self.load_image, [img], [tf.float32,tf.float32]),
72 | self.num_cpus)
73 |
74 | dataset = dataset.batch(self.batch_size)
75 |
76 | img1,img2 = dataset.make_one_shot_iterator().get_next()
77 |
78 | return img1,img2
79 |
--------------------------------------------------------------------------------
/tools/data_mean.py:
--------------------------------------------------------------------------------
1 |
2 | import cv2, argparse, os
3 | from glob import glob
4 | from tqdm import tqdm
5 |
6 | def parse_args():
7 | desc = "get the mean values of b,g,r on the whole dataset"
8 | parser = argparse.ArgumentParser(description=desc)
9 | parser.add_argument('--dataset', type=str, default='Hayao', help='dataset_name')
10 |
11 | return parser.parse_args()
12 |
13 | def read_img(image_path):
14 | img = cv2.imread(image_path)
15 | assert len(img.shape) == 3
16 | B = img[..., 0].mean()
17 | G = img[..., 1].mean()
18 | R = img[..., 2].mean()
19 | return B,G,R
20 |
21 | def get_mean(dataset_name):
22 | file_list = glob(os.path.dirname(os.path.dirname(__file__))+'/dataset/{}/{}/*.*'.format(dataset_name, 'style'))
23 | image_num = len(file_list)
24 | print('image_num:', image_num)
25 |
26 | B_total = 0
27 | G_total = 0
28 | R_total = 0
29 | for f in tqdm(file_list) :
30 | bgr = read_img(f)
31 | B_total += bgr[0]
32 | G_total += bgr[1]
33 | R_total += bgr[2]
34 |
35 | B_mean, G_mean, R_mean = B_total / image_num, G_total / image_num, R_total / image_num
36 | mean = (B_mean + G_mean + R_mean)/3
37 |
38 | return mean-B_mean, mean-G_mean, mean-R_mean
39 |
40 | """main"""
41 | def main():
42 | # parse arguments
43 | args = parse_args()
44 | if args is None:
45 | exit()
46 |
47 | return get_mean(args.dataset)
48 | if __name__ == '__main__':
49 | result = main()
50 | print('style_data_mean_diff (B, G, R): ', result)
--------------------------------------------------------------------------------
/tools/edge_smooth.py:
--------------------------------------------------------------------------------
1 | # The edge_smooth.py is from taki0112/CartoonGAN-Tensorflow https://github.com/taki0112/CartoonGAN-Tensorflow#2-do-edge_smooth
2 | from tools.utils import check_folder
3 | import numpy as np
4 | import cv2, os, argparse
5 | from glob import glob
6 | from tqdm import tqdm
7 |
8 | def parse_args():
9 | desc = "Edge smoothed"
10 | parser = argparse.ArgumentParser(description=desc)
11 | parser.add_argument('--dataset', type=str, default='Paprika', help='dataset_name')
12 | parser.add_argument('--img_size', type=int, default=256, help='The size of image')
13 |
14 | return parser.parse_args()
15 |
16 | def make_edge_smooth(dataset_name, img_size) :
17 | check_folder(os.path.dirname(os.path.dirname(__file__))+'/dataset/{}/{}'.format(dataset_name, 'smooth'))
18 |
19 | file_list = glob(os.path.dirname(os.path.dirname(__file__))+'/dataset/{}/{}/*.*'.format(dataset_name, 'style'))
20 | save_dir = os.path.dirname(os.path.dirname(__file__))+'/dataset/{}/smooth'.format(dataset_name)
21 |
22 | kernel_size = 5
23 | kernel = np.ones((kernel_size, kernel_size), np.uint8)
24 | gauss = cv2.getGaussianKernel(kernel_size, 0)
25 | gauss = gauss * gauss.transpose(1, 0)
26 |
27 | for f in tqdm(file_list) :
28 | file_name = os.path.basename(f)
29 |
30 | bgr_img = cv2.imread(f)
31 | gray_img = cv2.imread(f, 0)
32 |
33 | bgr_img = cv2.resize(bgr_img, (img_size, img_size))
34 | pad_img = np.pad(bgr_img, ((2, 2), (2, 2), (0, 0)), mode='reflect')
35 | gray_img = cv2.resize(gray_img, (img_size, img_size))
36 |
37 | edges = cv2.Canny(gray_img, 100, 200)
38 | dilation = cv2.dilate(edges, kernel)
39 |
40 | gauss_img = np.copy(bgr_img)
41 | idx = np.where(dilation != 0)
42 | for i in range(np.sum(dilation != 0)):
43 | gauss_img[idx[0][i], idx[1][i], 0] = np.sum(
44 | np.multiply(pad_img[idx[0][i]:idx[0][i] + kernel_size, idx[1][i]:idx[1][i] + kernel_size, 0], gauss))
45 | gauss_img[idx[0][i], idx[1][i], 1] = np.sum(
46 | np.multiply(pad_img[idx[0][i]:idx[0][i] + kernel_size, idx[1][i]:idx[1][i] + kernel_size, 1], gauss))
47 | gauss_img[idx[0][i], idx[1][i], 2] = np.sum(
48 | np.multiply(pad_img[idx[0][i]:idx[0][i] + kernel_size, idx[1][i]:idx[1][i] + kernel_size, 2], gauss))
49 |
50 | cv2.imwrite(os.path.join(save_dir, file_name), gauss_img)
51 |
52 | """main"""
53 | def main():
54 | # parse arguments
55 | args = parse_args()
56 | if args is None:
57 | exit()
58 |
59 | make_edge_smooth(args.dataset, args.img_size)
60 |
61 |
62 | if __name__ == '__main__':
63 | main()
64 |
--------------------------------------------------------------------------------
/tools/get_generator_ckpt.py:
--------------------------------------------------------------------------------
1 | import argparse
2 | from tools.utils import *
3 | import os
4 | from net import generator
5 |
6 | os.environ["CUDA_VISIBLE_DEVICES"] = "0"
7 |
8 | def parse_args():
9 | desc = "AnimeGAN"
10 | parser = argparse.ArgumentParser(description=desc)
11 |
12 | parser.add_argument('--checkpoint_dir', type=str, default='../checkpoint/' + 'AnimeGAN_Hayao_lsgan_300_300_1_2_10',
13 | help='Directory name to save the checkpoints')
14 | parser.add_argument('--style_name', type=str, default='Hayao',
15 | help='what style you want to get')
16 |
17 | return parser.parse_args()
18 |
19 | def save(saver, sess, checkpoint_dir, model_name):
20 |
21 | save_path = os.path.join(checkpoint_dir, model_name + '.ckpt')
22 | saver.save(sess, save_path, write_meta_graph=True)
23 | return save_path
24 |
25 | def main(checkpoint_dir, style_name, ):
26 |
27 | ckpt_dir = '../checkpoint/' + 'generator_'+style_name+'_weight'
28 | check_folder(ckpt_dir)
29 |
30 | placeholder = tf.placeholder(tf.float32, [1, None, None, 3], name='generator_input')
31 | with tf.variable_scope("generator", reuse=False):
32 | _ = generator.G_net(placeholder).fake
33 |
34 | generator_var = [var for var in tf.trainable_variables() if var.name.startswith('generator')]
35 | saver = tf.train.Saver(generator_var)
36 |
37 | gpu_options = tf.GPUOptions(allow_growth=True)
38 | with tf.Session(config=tf.ConfigProto(allow_soft_placement=True, gpu_options=gpu_options)) as sess:
39 | sess.run(tf.global_variables_initializer())
40 | # load model
41 | ckpt = tf.train.get_checkpoint_state(checkpoint_dir) # checkpoint file information
42 | if ckpt and ckpt.model_checkpoint_path:
43 | print(ckpt.model_checkpoint_path)
44 | ckpt_name = os.path.basename(ckpt.model_checkpoint_path) # first line
45 | saver.restore(sess, os.path.join(checkpoint_dir, ckpt_name))
46 | counter = ckpt_name.split('-')[-1]
47 | print(" [*] Success to read {}".format(ckpt_name))
48 | else:
49 | print(" [*] Failed to find a checkpoint")
50 | return
51 |
52 | info = save(saver, sess, ckpt_dir, style_name+'-'+counter)
53 |
54 | print(f'save over : {info} ')
55 |
56 |
57 |
58 | if __name__ == '__main__':
59 | arg = parse_args()
60 | print(arg.checkpoint_dir)
61 | main(arg.checkpoint_dir, arg.style_name)
62 |
--------------------------------------------------------------------------------
/tools/ops.py:
--------------------------------------------------------------------------------
1 | import tensorflow as tf
2 | import tensorflow.contrib as tf_contrib
3 |
4 |
5 | # Xavier : tf_contrib.layers.xavier_initializer()
6 | # He : tf_contrib.layers.variance_scaling_initializer()
7 | # Normal : tf.random_normal_initializer(mean=0.0, stddev=0.02)
8 | # l2_decay : tf_contrib.layers.l2_regularizer(0.0001)
9 |
10 |
11 | weight_init = tf.random_normal_initializer(mean=0.0, stddev=0.02)
12 | weight_regularizer = None
13 |
14 | ##################################################################################
15 | # Layer
16 | ##################################################################################
17 |
18 | def conv(x, channels, kernel=4, stride=2, pad=0, pad_type='zero', use_bias=True, sn=False, scope='conv_0'):
19 | with tf.variable_scope(scope):
20 | if (kernel - stride) % 2 == 0 :
21 | pad_top = pad
22 | pad_bottom = pad
23 | pad_left = pad
24 | pad_right = pad
25 |
26 | else :
27 | pad_top = pad
28 | pad_bottom = kernel - stride - pad_top
29 | pad_left = pad
30 | pad_right = kernel - stride - pad_left
31 |
32 | if pad_type == 'zero' :
33 | x = tf.pad(x, [[0, 0], [pad_top, pad_bottom], [pad_left, pad_right], [0, 0]])
34 | if pad_type == 'reflect' :
35 | x = tf.pad(x, [[0, 0], [pad_top, pad_bottom], [pad_left, pad_right], [0, 0]], mode='REFLECT')
36 |
37 | if sn :
38 | w = tf.get_variable("kernel", shape=[kernel, kernel, x.get_shape()[-1], channels], initializer=weight_init,
39 | regularizer=weight_regularizer)
40 | x = tf.nn.conv2d(input=x, filter=spectral_norm(w),
41 | strides=[1, stride, stride, 1], padding='VALID')
42 | if use_bias :
43 | bias = tf.get_variable("bias", [channels], initializer=tf.constant_initializer(0.0))
44 | x = tf.nn.bias_add(x, bias)
45 |
46 | else :
47 | x = tf.layers.conv2d(inputs=x, filters=channels,
48 | kernel_size=kernel, kernel_initializer=weight_init,
49 | kernel_regularizer=weight_regularizer,
50 | strides=stride, use_bias=use_bias)
51 |
52 |
53 | return x
54 |
55 | def deconv(x, channels, kernel=4, stride=2, use_bias=True, sn=False, scope='deconv_0'):
56 | with tf.variable_scope(scope):
57 | x_shape = x.get_shape().as_list()
58 | output_shape = [x_shape[0], tf.shape(x)[1]*stride, tf.shape(x)[2]*stride, channels]
59 | if sn :
60 | w = tf.get_variable("kernel", shape=[kernel, kernel, channels, x.get_shape()[-1]], initializer=weight_init, regularizer=weight_regularizer)
61 | x = tf.nn.conv2d_transpose(x, filter=spectral_norm(w), output_shape=output_shape, strides=[1, stride, stride, 1], padding='SAME')
62 |
63 | if use_bias :
64 | bias = tf.get_variable("bias", [channels], initializer=tf.constant_initializer(0.0))
65 | x = tf.nn.bias_add(x, bias)
66 |
67 | else :
68 | x = tf.layers.conv2d_transpose(inputs=x, filters=channels,
69 | kernel_size=kernel, kernel_initializer=weight_init, kernel_regularizer=weight_regularizer,
70 | strides=stride, padding='SAME', use_bias=use_bias)
71 |
72 | return x
73 |
74 |
75 | ##################################################################################
76 | # Residual-block
77 | ##################################################################################
78 |
79 | def resblock(x_init, channels, use_bias=True, scope='resblock_0'):
80 | with tf.variable_scope(scope):
81 | with tf.variable_scope('res1'):
82 | x = conv(x_init, channels, kernel=3, stride=1, pad=1, pad_type='reflect', use_bias=use_bias)
83 | x = instance_norm(x)
84 | x = relu(x)
85 |
86 | with tf.variable_scope('res2'):
87 | x = conv(x, channels, kernel=3, stride=1, pad=1, pad_type='reflect', use_bias=use_bias)
88 | x = instance_norm(x)
89 |
90 | return x + x_init
91 |
92 | ##################################################################################
93 | # Sampling
94 | ##################################################################################
95 |
96 | def flatten(x) :
97 | return tf.layers.flatten(x)
98 |
99 | ##################################################################################
100 | # Activation function
101 | ##################################################################################
102 |
103 | def lrelu(x, alpha=0.2):
104 | return tf.nn.leaky_relu(x, alpha)
105 |
106 |
107 | def relu(x):
108 | return tf.nn.relu(x)
109 |
110 |
111 | def tanh(x):
112 | return tf.tanh(x)
113 |
114 | def sigmoid(x) :
115 | return tf.sigmoid(x)
116 |
117 | ##################################################################################
118 | # Normalization function
119 | ##################################################################################
120 |
121 | def instance_norm(x, scope='instance_norm'):
122 | return tf_contrib.layers.instance_norm(x,
123 | epsilon=1e-05,
124 | center=True, scale=True,
125 | scope=scope)
126 |
127 | def layer_norm(x, scope='layer_norm') :
128 | return tf_contrib.layers.layer_norm(x,
129 | center=True, scale=True,
130 | scope=scope)
131 |
132 | def batch_norm(x, is_training=True, scope='batch_norm'):
133 | return tf_contrib.layers.batch_norm(x,
134 | decay=0.9, epsilon=1e-05,
135 | center=True, scale=True, updates_collections=None,
136 | is_training=is_training, scope=scope)
137 |
138 |
139 | def spectral_norm(w, iteration=1):
140 | w_shape = w.shape.as_list()
141 | w = tf.reshape(w, [-1, w_shape[-1]])
142 |
143 | u = tf.get_variable("u", [1, w_shape[-1]], initializer=tf.truncated_normal_initializer(), trainable=False)
144 |
145 | u_hat = u
146 | v_hat = None
147 | for i in range(iteration):
148 | """
149 | power iteration
150 | Usually iteration = 1 will be enough
151 | """
152 | v_ = tf.matmul(u_hat, tf.transpose(w))
153 | v_hat = l2_norm(v_)
154 |
155 | u_ = tf.matmul(v_hat, w)
156 | u_hat = l2_norm(u_)
157 |
158 | sigma = tf.matmul(tf.matmul(v_hat, w), tf.transpose(u_hat))
159 | w_norm = w / sigma
160 |
161 | with tf.control_dependencies([u.assign(u_hat)]):
162 | w_norm = tf.reshape(w_norm, w_shape)
163 |
164 | return w_norm
165 |
166 | def l2_norm(v, eps=1e-12):
167 | return v / (tf.reduce_sum(v ** 2) ** 0.5 + eps)
168 |
169 | ##################################################################################
170 | # Loss function
171 | ##################################################################################
172 |
173 | def L1_loss(x, y):
174 | loss = tf.reduce_mean(tf.abs(x - y))
175 | return loss
176 |
177 | def L2_loss(x,y):
178 | size = tf.size(x)
179 | return tf.nn.l2_loss(x-y)* 2 / tf.to_float(size)
180 |
181 | def Huber_loss(x,y):
182 | return tf.losses.huber_loss(x,y)
183 |
184 | def discriminator_loss(loss_func, real, gray, fake, real_blur):
185 | real_loss = 0
186 | gray_loss = 0
187 | fake_loss = 0
188 | real_blur_loss = 0
189 |
190 |
191 | if loss_func == 'wgan-gp' or loss_func == 'wgan-lp':
192 | real_loss = -tf.reduce_mean(real)
193 | gray_loss = tf.reduce_mean(gray)
194 | fake_loss = tf.reduce_mean(fake)
195 | real_blur_loss = tf.reduce_mean(real_blur)
196 |
197 | if loss_func == 'lsgan' :
198 | real_loss = tf.reduce_mean(tf.square(real - 1.0))
199 | gray_loss = tf.reduce_mean(tf.square(gray))
200 | fake_loss = tf.reduce_mean(tf.square(fake))
201 | real_blur_loss = tf.reduce_mean(tf.square(real_blur))
202 |
203 | if loss_func == 'gan' or loss_func == 'dragan' :
204 | real_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(real), logits=real))
205 | gray_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.zeros_like(gray), logits=gray))
206 | fake_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.zeros_like(fake), logits=fake))
207 | real_blur_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.zeros_like(real_blur), logits=real_blur))
208 |
209 | if loss_func == 'hinge':
210 | real_loss = tf.reduce_mean(relu(1.0 - real))
211 | gray_loss = tf.reduce_mean(relu(1.0 + gray))
212 | fake_loss = tf.reduce_mean(relu(1.0 + fake))
213 | real_blur_loss = tf.reduce_mean(relu(1.0 + real_blur))
214 |
215 | loss = real_loss + fake_loss + gray_loss + 0.1 * real_blur_loss
216 |
217 | return loss
218 |
219 | def generator_loss(loss_func, fake):
220 | fake_loss = 0
221 |
222 | if loss_func == 'wgan-gp' or loss_func == 'wgan-lp':
223 | fake_loss = -tf.reduce_mean(fake)
224 |
225 | if loss_func == 'lsgan' :
226 | fake_loss = tf.reduce_mean(tf.square(fake - 1.0))
227 |
228 | if loss_func == 'gan' or loss_func == 'dragan':
229 | fake_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(fake), logits=fake))
230 |
231 | if loss_func == 'hinge':
232 | fake_loss = -tf.reduce_mean(fake)
233 |
234 | loss = fake_loss
235 |
236 | return loss
237 |
238 | def gram(x):
239 | shape_x = tf.shape(x)
240 | b = shape_x[0]
241 | c = shape_x[3]
242 | x = tf.reshape(x, [b, -1, c])
243 | return tf.matmul(tf.transpose(x, [0, 2, 1]), x) / tf.cast((tf.size(x) // b), tf.float32)
244 |
245 | def con_loss(vgg, real, fake):
246 |
247 | vgg.build(real)
248 | real_feature_map = vgg.conv4_4_no_activation
249 |
250 | vgg.build(fake)
251 | fake_feature_map = vgg.conv4_4_no_activation
252 |
253 | loss = L1_loss(real_feature_map, fake_feature_map)
254 |
255 | return loss
256 |
257 |
258 | def style_loss(style, fake):
259 | return L1_loss(gram(style), gram(fake))
260 |
261 | def con_sty_loss(vgg, real, anime, fake):
262 |
263 | vgg.build(real)
264 | real_feature_map = vgg.conv4_4_no_activation
265 |
266 | vgg.build(fake)
267 | fake_feature_map = vgg.conv4_4_no_activation
268 |
269 | vgg.build(anime[:fake_feature_map.shape[0]])
270 | anime_feature_map = vgg.conv4_4_no_activation
271 |
272 | c_loss = L1_loss(real_feature_map, fake_feature_map)
273 | s_loss = style_loss(anime_feature_map, fake_feature_map)
274 |
275 | return c_loss,s_loss
276 |
277 | def color_loss(con, fake):
278 | con = rgb2yuv(con)
279 | fake = rgb2yuv(fake)
280 |
281 | return L1_loss(con[:,:,:,0], fake[:,:,:,0]) + Huber_loss(con[:,:,:,1],fake[:,:,:,1]) + Huber_loss(con[:,:,:,2],fake[:,:,:,2])
282 |
283 | def rgb2yuv(rgb):
284 | """
285 | Convert RGB image into YUV https://en.wikipedia.org/wiki/YUV
286 | """
287 | rgb = (rgb + 1.0)/2.0
288 | # rgb2yuv_filter = tf.constant([[[[0.299, -0.169, 0.499],
289 | # [0.587, -0.331, -0.418],
290 | # [0.114, 0.499, -0.0813]]]])
291 | # rgb2yuv_bias = tf.constant([0., 0.5, 0.5])
292 | # temp = tf.nn.conv2d(rgb, rgb2yuv_filter, [1, 1, 1, 1], 'SAME')
293 | # temp = tf.nn.bias_add(temp, rgb2yuv_bias)
294 | # return temp
295 | return tf.image.rgb_to_yuv(rgb)
296 |
297 |
--------------------------------------------------------------------------------
/tools/utils.py:
--------------------------------------------------------------------------------
1 | import tensorflow as tf
2 | from tensorflow.contrib import slim
3 | from tools.adjust_brightness import adjust_brightness_from_src_to_dst, read_img
4 | import os,cv2
5 | import numpy as np
6 |
7 |
8 | def load_test_data(image_path, size):
9 | img = cv2.imread(image_path).astype(np.float32)
10 | img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
11 | img = preprocessing(img,size)
12 | img = np.expand_dims(img, axis=0)
13 | return img
14 |
15 | def preprocessing(img, size):
16 | h, w = img.shape[:2]
17 | if h <= size[0]:
18 | h = size[0]
19 | else:
20 | x = h % 32
21 | h = h - x
22 |
23 | if w < size[1]:
24 | w = size[1]
25 | else:
26 | y = w % 32
27 | w = w - y
28 | # the cv2 resize func : dsize format is (W ,H)
29 | img = cv2.resize(img, (w, h))
30 | return img/127.5 - 1.0
31 |
32 |
33 | def save_images(images, image_path, photo_path = None):
34 | fake = inverse_transform(images.squeeze())
35 | if photo_path:
36 | return imsave(adjust_brightness_from_src_to_dst(fake, read_img(photo_path)), image_path)
37 | else:
38 | return imsave(fake, image_path)
39 |
40 | def inverse_transform(images):
41 | images = (images + 1.) / 2 * 255
42 | # The calculation of floating-point numbers is inaccurate,
43 | # and the range of pixel values must be limited to the boundary,
44 | # otherwise, image distortion or artifacts will appear during display.
45 | images = np.clip(images, 0, 255)
46 | return images.astype(np.uint8)
47 |
48 |
49 | def imsave(images, path):
50 | return cv2.imwrite(path, cv2.cvtColor(images, cv2.COLOR_BGR2RGB))
51 |
52 | crop_image = lambda img, x0, y0, w, h: img[y0:y0+h, x0:x0+w]
53 |
54 | def random_crop(img1, img2, crop_H, crop_W):
55 |
56 | assert img1.shape == img2.shape
57 | h, w = img1.shape[:2]
58 |
59 | # The crop width cannot exceed the original image crop width
60 | if crop_W > w:
61 | crop_W = w
62 |
63 | # Crop height
64 | if crop_H > h:
65 | crop_H = h
66 |
67 | # Randomly generate the position of the upper left corner
68 | x0 = np.random.randint(0, w - crop_W + 1)
69 | y0 = np.random.randint(0, h - crop_H + 1)
70 |
71 | crop_1 = crop_image(img1, x0, y0, crop_W, crop_H)
72 | crop_2 = crop_image(img2, x0, y0, crop_W, crop_H)
73 | return crop_1,crop_2
74 |
75 |
76 | def show_all_variables():
77 | model_vars = tf.trainable_variables()
78 |
79 | slim.model_analyzer.analyze_vars(model_vars, print_info=True)
80 | print('G:')
81 | slim.model_analyzer.analyze_vars([var for var in tf.trainable_variables() if var.name.startswith('generator')], print_info=True)
82 | # print('D:')
83 | # slim.model_analyzer.analyze_vars([var for var in tf.trainable_variables() if var.name.startswith('discriminator')], print_info=True)
84 |
85 | def check_folder(log_dir):
86 | if not os.path.exists(log_dir):
87 | os.makedirs(log_dir)
88 | return log_dir
89 |
90 | def str2bool(x):
91 | return x.lower() in ('true')
92 |
93 |
94 |
--------------------------------------------------------------------------------
/tools/vgg19.py:
--------------------------------------------------------------------------------
1 |
2 | import tensorflow as tf
3 |
4 | import numpy as np
5 | import time
6 | import sys
7 |
8 | VGG_MEAN = [103.939, 116.779, 123.68]
9 |
10 |
11 | class Vgg19:
12 | def __init__(self, vgg19_npy_path='vgg19_weight/vgg19.npy'):
13 |
14 | if vgg19_npy_path is not None:
15 | self.data_dict = np.load(vgg19_npy_path, encoding='latin1', allow_pickle=True).item()
16 | print("npy file loaded ------- ",vgg19_npy_path)
17 | else:
18 | self.data_dict = None
19 | print("npy file load error!")
20 | sys.exit(1)
21 |
22 | def build(self, rgb, include_fc=False):
23 | """
24 | load variable from npy to build the VGG
25 | input format: bgr image with shape [batch_size, h, w, 3]
26 | scale: (-1, 1)
27 | """
28 |
29 | start_time = time.time()
30 | rgb_scaled = ((rgb + 1) / 2) * 255.0 # [-1, 1] ~ [0, 255]
31 |
32 | red, green, blue = tf.split(axis=3, num_or_size_splits=3, value=rgb_scaled)
33 | bgr = tf.concat(axis=3, values=[blue - VGG_MEAN[0],
34 | green - VGG_MEAN[1],
35 | red - VGG_MEAN[2]])
36 |
37 | self.conv1_1 = self.conv_layer(bgr, "conv1_1")
38 | self.conv1_2 = self.conv_layer(self.conv1_1, "conv1_2")
39 | self.pool1 = self.max_pool(self.conv1_2, 'pool1')
40 |
41 | self.conv2_1 = self.conv_layer(self.pool1, "conv2_1")
42 | self.conv2_2 = self.conv_layer(self.conv2_1, "conv2_2")
43 | self.pool2 = self.max_pool(self.conv2_2, 'pool2')
44 |
45 | self.conv3_1 = self.conv_layer(self.pool2, "conv3_1")
46 | self.conv3_2 = self.conv_layer(self.conv3_1, "conv3_2")
47 | self.conv3_3 = self.conv_layer(self.conv3_2, "conv3_3")
48 | self.conv3_4 = self.conv_layer(self.conv3_3, "conv3_4")
49 | self.pool3 = self.max_pool(self.conv3_4, 'pool3')
50 |
51 | self.conv4_1 = self.conv_layer(self.pool3, "conv4_1")
52 | self.conv4_2 = self.conv_layer(self.conv4_1, "conv4_2")
53 | self.conv4_3 = self.conv_layer(self.conv4_2, "conv4_3")
54 |
55 | self.conv4_4_no_activation = self.no_activation_conv_layer(self.conv4_3, "conv4_4")
56 |
57 | self.conv4_4 = self.conv_layer(self.conv4_3, "conv4_4")
58 | self.pool4 = self.max_pool(self.conv4_4, 'pool4')
59 |
60 | self.conv5_1 = self.conv_layer(self.pool4, "conv5_1")
61 | self.conv5_2 = self.conv_layer(self.conv5_1, "conv5_2")
62 | self.conv5_3 = self.conv_layer(self.conv5_2, "conv5_3")
63 | self.conv5_4 = self.conv_layer(self.conv5_3, "conv5_4")
64 | self.pool5 = self.max_pool(self.conv5_4, 'pool5')
65 |
66 | if include_fc:
67 | self.fc6 = self.fc_layer(self.pool5, "fc6")
68 | assert self.fc6.get_shape().as_list()[1:] == [4096]
69 | self.relu6 = tf.nn.relu(self.fc6)
70 |
71 | self.fc7 = self.fc_layer(self.relu6, "fc7")
72 | self.relu7 = tf.nn.relu(self.fc7)
73 |
74 | self.fc8 = self.fc_layer(self.relu7, "fc8")
75 |
76 | self.prob = tf.nn.softmax(self.fc8, name="prob")
77 |
78 | self.data_dict = None
79 | print(("build model finished: %fs" % (time.time() - start_time)))
80 |
81 | def avg_pool(self, bottom, name):
82 | return tf.nn.avg_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)
83 |
84 | def max_pool(self, bottom, name):
85 | return tf.nn.max_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)
86 |
87 | def conv_layer(self, bottom, name):
88 | with tf.variable_scope(name):
89 | filt = self.get_conv_filter(name)
90 |
91 | conv = tf.nn.conv2d(bottom, filt, [1, 1, 1, 1], padding='SAME')
92 |
93 | conv_biases = self.get_bias(name)
94 | bias = tf.nn.bias_add(conv, conv_biases)
95 |
96 | relu = tf.nn.relu(bias)
97 | return relu
98 |
99 | def no_activation_conv_layer(self, bottom, name):
100 | with tf.variable_scope(name):
101 | filt = self.get_conv_filter(name)
102 |
103 | conv = tf.nn.conv2d(bottom, filt, [1, 1, 1, 1], padding='SAME')
104 |
105 | conv_biases = self.get_bias(name)
106 | x = tf.nn.bias_add(conv, conv_biases)
107 |
108 | return x
109 |
110 | def fc_layer(self, bottom, name):
111 | with tf.variable_scope(name):
112 | shape = bottom.get_shape().as_list()
113 | dim = 1
114 | for d in shape[1:]:
115 | dim *= d
116 | x = tf.reshape(bottom, [-1, dim])
117 |
118 | weights = self.get_fc_weight(name)
119 | biases = self.get_bias(name)
120 |
121 | # Fully connected layer. Note that the '+' operation automatically
122 | # broadcasts the biases.
123 | fc = tf.nn.bias_add(tf.matmul(x, weights), biases)
124 |
125 | return fc
126 |
127 | def get_conv_filter(self, name):
128 | return tf.constant(self.data_dict[name][0], name="filter")
129 |
130 | def get_bias(self, name):
131 | return tf.constant(self.data_dict[name][1], name="biases")
132 |
133 | def get_fc_weight(self, name):
134 | return tf.constant(self.data_dict[name][0], name="weights")
--------------------------------------------------------------------------------
/train.py:
--------------------------------------------------------------------------------
1 | from AnimeGAN import AnimeGAN
2 | import argparse
3 | from tools.utils import *
4 | import os
5 | os.environ["CUDA_VISIBLE_DEVICES"] = "1"
6 |
7 | """parsing and configuration"""
8 |
9 | def parse_args():
10 | desc = "Tensorflow implementation of AnimeGAN"
11 | parser = argparse.ArgumentParser(description=desc)
12 | parser.add_argument('--dataset', type=str, default='Hayao', help='dataset_name')
13 |
14 | parser.add_argument('--epoch', type=int, default=101, help='The number of epochs to run')
15 | parser.add_argument('--init_epoch', type=int, default=5, help='The number of epochs for weight initialization')
16 | parser.add_argument('--batch_size', type=int, default=6, help='The size of batch size')
17 | parser.add_argument('--save_freq', type=int, default=1, help='The number of ckpt_save_freq')
18 |
19 | parser.add_argument('--init_lr', type=float, default=1e-4, help='The learning rate')
20 | parser.add_argument('--g_lr', type=float, default=8e-5, help='The learning rate')
21 | parser.add_argument('--d_lr', type=float, default=16e-5, help='The learning rate')
22 | parser.add_argument('--ld', type=float, default=10.0, help='The gradient penalty lambda')
23 |
24 | parser.add_argument('--g_adv_weight', type=float, default=300.0, help='Weight about GAN')
25 | parser.add_argument('--d_adv_weight', type=float, default=300.0, help='Weight about GAN')
26 | parser.add_argument('--con_weight', type=float, default=1.5, help='Weight about VGG19') # 1.1 for Shinkai
27 | # ------ the follow weight used in AnimeGAN
28 | parser.add_argument('--sty_weight', type=float, default=3.0, help='Weight about style')
29 | parser.add_argument('--color_weight', type=float, default=10.0, help='Weight about color')
30 | # ---------------------------------------------
31 | parser.add_argument('--training_rate', type=int, default=1, help='training rate about G & D')
32 | parser.add_argument('--gan_type', type=str, default='lsgan', help='[gan / lsgan / wgan-gp / wgan-lp / dragan / hinge')
33 |
34 | parser.add_argument('--img_size', type=list, default=[256,256], help='The size of image: H and W')
35 | parser.add_argument('--img_ch', type=int, default=3, help='The size of image channel')
36 |
37 | parser.add_argument('--ch', type=int, default=64, help='base channel number per layer')
38 | parser.add_argument('--n_dis', type=int, default=3, help='The number of discriminator layer')
39 | parser.add_argument('--sn', type=str2bool, default=True, help='using spectral norm')
40 |
41 | parser.add_argument('--checkpoint_dir', type=str, default='checkpoint',
42 | help='Directory name to save the checkpoints')
43 | parser.add_argument('--log_dir', type=str, default='logs',
44 | help='Directory name to save training logs')
45 | parser.add_argument('--sample_dir', type=str, default='samples',
46 | help='Directory name to save the samples on training')
47 |
48 | return check_args(parser.parse_args())
49 |
50 | """checking arguments"""
51 | def check_args(args):
52 | # --checkpoint_dir
53 | check_folder(args.checkpoint_dir)
54 |
55 | # --log_dir
56 | check_folder(args.log_dir)
57 |
58 | # --sample_dir
59 | check_folder(args.sample_dir)
60 |
61 | # --epoch
62 | try:
63 | assert args.epoch >= 1
64 | except:
65 | print('number of epochs must be larger than or equal to one')
66 |
67 | # --batch_size
68 | try:
69 | assert args.batch_size >= 1
70 | except:
71 | print('batch size must be larger than or equal to one')
72 | return args
73 |
74 |
75 | """main"""
76 | def main():
77 | # parse arguments
78 | args = parse_args()
79 | if args is None:
80 | exit()
81 |
82 | # open session
83 | gpu_options = tf.GPUOptions(allow_growth=True)
84 | with tf.Session(config=tf.ConfigProto(allow_soft_placement=True,inter_op_parallelism_threads=8,
85 | intra_op_parallelism_threads=8,gpu_options=gpu_options)) as sess:
86 | gan = AnimeGAN(sess, args)
87 |
88 | # build graph
89 | gan.build_model()
90 |
91 | # show network architecture
92 | show_all_variables()
93 |
94 | gan.train()
95 | print(" [*] Training finished!")
96 |
97 |
98 |
99 | if __name__ == '__main__':
100 | main()
101 |
--------------------------------------------------------------------------------
/vgg19_weight/vgg19.npy file:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/video/input/お花見.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/video/input/お花見.mp4
--------------------------------------------------------------------------------
/video/output/お花見.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/TachibanaYoshino/AnimeGAN/b7c16f7c58e6c609f4064b1228782419a7e7a45c/video/output/お花見.mp4
--------------------------------------------------------------------------------
/video2anime.py:
--------------------------------------------------------------------------------
1 | '''
2 | made by @finnkso (github)
3 | 2020.04.09
4 | '''
5 | import argparse
6 | import os
7 | import cv2
8 | from tqdm import tqdm
9 | import numpy as np
10 | import tensorflow as tf
11 |
12 | from net import generator
13 | from tools.utils import preprocessing, check_folder
14 | from tools.adjust_brightness import adjust_brightness_from_src_to_dst
15 |
16 | os.environ["CUDA_VISIBLE_DEVICES"] = "0"
17 |
18 | def parse_args():
19 | desc = "Tensorflow implementation of AnimeGAN"
20 | parser = argparse.ArgumentParser(description=desc)
21 | parser.add_argument('--video', type=str, default='video/input/'+ 'お花見.mp4',
22 | help='video file or number for webcam')
23 | parser.add_argument('--checkpoint_dir', type=str, default='../checkpoint/generator_Hayao_weight',
24 | help='Directory name to save the checkpoints')
25 | parser.add_argument('--output', type=str, default='video/output',
26 | help='output path')
27 | parser.add_argument('--output_format', type=str, default='MP4V',
28 | help='codec used in VideoWriter when saving video to file')
29 | parser.add_argument('--if_adjust_brightness', type=bool, default=False,
30 | help='adjust brightness by the real photo')
31 | return parser.parse_args()
32 |
33 |
34 | def convert_image(img, img_size):
35 | img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
36 | img = preprocessing(img, img_size)
37 | img = np.expand_dims(img, axis=0)
38 | img = np.asarray(img)
39 | return img
40 |
41 | def inverse_image(img):
42 | img = (img.squeeze()+1.) / 2 * 255
43 | img = img.astype(np.uint8)
44 | return img
45 |
46 | def cvt2anime_video(video, output, checkpoint_dir, output_format='MP4V', if_adjust_brightness=False, img_size=(256,256)):
47 | '''
48 | output_format: 4-letter code that specify codec to use for specific video type. e.g. for mp4 support use "H264", "MP4V", or "X264"
49 | '''
50 | # tf.reset_default_graph()
51 | # check_folder(result_dir)
52 | gpu_stat = bool(len(tf.config.experimental.list_physical_devices('GPU')))
53 | if gpu_stat:
54 | os.environ["CUDA_VISIBLE_DEVICES"] = "0"
55 | gpu_options = tf.GPUOptions(allow_growth=gpu_stat)
56 |
57 | test_real = tf.placeholder(tf.float32, [1, None, None, 3], name='test')
58 |
59 | with tf.variable_scope("generator", reuse=False):
60 | test_generated = generator.G_net(test_real).fake
61 |
62 | # load video
63 | vid = cv2.VideoCapture(video)
64 | vid_name = os.path.basename(video)
65 | total = int(vid.get(cv2.CAP_PROP_FRAME_COUNT))
66 | fps = vid.get(cv2.CAP_PROP_FPS)
67 | # codec = cv2.VideoWriter_fourcc('M', 'J', 'P', 'G')
68 | codec = cv2.VideoWriter_fourcc(*output_format)
69 |
70 | tfconfig = tf.ConfigProto(allow_soft_placement=True, gpu_options=gpu_options)
71 | with tf.Session(config=tfconfig) as sess:
72 | # tf.global_variables_initializer().run()
73 | # load model
74 | ckpt = tf.train.get_checkpoint_state(checkpoint_dir) # checkpoint file information
75 | saver = tf.train.Saver()
76 | if ckpt and ckpt.model_checkpoint_path:
77 | ckpt_name = os.path.basename(ckpt.model_checkpoint_path) # first line
78 | saver.restore(sess, os.path.join(checkpoint_dir, ckpt_name))
79 | print(" [*] Success to read {}".format(ckpt_name))
80 | else:
81 | print(" [*] Failed to find a checkpoint")
82 | return
83 |
84 | # determine output width and height
85 | ret, img = vid.read()
86 | if img is None:
87 | print('Error! Failed to determine frame size: frame empty.')
88 | return
89 | img = preprocessing(img, img_size)
90 | height, width = img.shape[:2]
91 | out = cv2.VideoWriter(os.path.join(output, vid_name), codec, fps, (width, height))
92 |
93 | pbar = tqdm(total=total)
94 | vid.set(cv2.CAP_PROP_POS_FRAMES, 0)
95 | while ret:
96 | ret, frame = vid.read()
97 | if frame is None:
98 | print('Warning: got empty frame.')
99 | continue
100 |
101 | img = convert_image(frame, img_size)
102 | fake_img = sess.run(test_generated, feed_dict={test_real: img})
103 | fake_img = inverse_image(fake_img)
104 | if if_adjust_brightness:
105 | fake_img = cv2.cvtColor(adjust_brightness_from_src_to_dst(fake_img, frame), cv2.COLOR_BGR2RGB)
106 | else:
107 | fake_img = cv2.cvtColor(fake_img, cv2.COLOR_BGR2RGB)
108 | fake_img = cv2.resize(fake_img, (width, height))
109 | out.write(fake_img)
110 | pbar.update(1)
111 |
112 | pbar.close()
113 | vid.release()
114 | # cv2.destroyAllWindows()
115 | return os.path.join(output, vid_name)
116 |
117 |
118 | if __name__ == '__main__':
119 | arg = parse_args()
120 | check_folder(arg.output)
121 | info = cvt2anime_video(arg.video, arg.output, arg.checkpoint_dir, output_format=arg.output_format, if_adjust_brightness=arg.if_adjust_brightness)
122 | print(f'output video: {info}')
123 |
--------------------------------------------------------------------------------