File size: 22,427 Bytes
83034b6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 |
import torch
import models.networks as networks
import util.util as util
import util.logging_wandb as wblogging
import os
import numpy as np
import torch.nn.functional as F
from tqdm import tqdm
from torchvision.utils import save_image
import torch.nn as nn
import torchvision.transforms as transforms
import torch.utils.data as data_torch
from torch.utils.data import Dataset
from .sampler import InfiniteSamplerWrapper,save_checkpoint
from .networks.generator_module.schedule import CosineAnnealingWarmUpLR
import torch.utils.data as data
from rich.progress import Progress, TimeRemainingColumn, BarColumn, TextColumn
import datetime
class Pix2PixModel(torch.nn.Module):
@staticmethod
def modify_commandline_options(parser, is_train):
networks.modify_commandline_options(parser, is_train)
parser.add_argument('--lrG_decay', type=float, default=1e-4, help='learning rate decay')
parser.add_argument('--lrG', type=float, default=12e-4, help='initial learning rate for adam')
parser.add_argument('--save_dir', default=r'/home/share/VAL_ImgTranslations/experiments',
help='Directory to save the model')
parser.add_argument('--content_weight', type=float, default=2.0, help='weight for content reconstruction loss')
parser.add_argument('--save_model_interval', type=int, default=300)
parser.add_argument('--style_weight', type=float, default=3.0, help='weight for style reconstruction loss')
parser.add_argument('--max_iter', type=int, default=700)
parser.add_argument('--n_threads', type=int, default=1)
parser.add_argument('--batch_size', type=int, default=1)
parser.add_argument('--id1_weight', type=float, default=50)
parser.add_argument('--id2_weight', type=float, default=1)
parser.add_argument('--loss_count_interval', type=int, default=200)
return parser
def __init__(self, opt):
super().__init__()
self.opt = opt
self.loss=0
self.out=None
self.content_weight=opt.content_weight
self.style_weight=opt.style_weight
self.lrG_decay=opt.lrG_decay
self.lrG=opt.lrG
self.loss_count_interval=opt.loss_count_interval
self.save_model_interval=opt.save_model_interval
self.n_threads=opt.n_threads
self.save_dir=opt.save_dir
self.id1_weight=opt.id1_weight
self.id2_weight=opt.id2_weight
self.max_iter=opt.max_iter
self.epoch=0
self.batch_size=opt.batch_size
self.FloatTensor = torch.cuda.FloatTensor if self.use_gpu() \
else torch.FloatTensor
self.ByteTensor = torch.cuda.ByteTensor if self.use_gpu() \
else torch.ByteTensor
self.netG, self.netD,self.netD2, self.netE = self.initialize_networks(opt)
self.device="cuda" if torch.cuda.is_available() else "cpu"
# set loss functions
if opt.isTrain:
self.criterionGAN = networks.GANLoss(
opt.gan_mode, tensor=self.FloatTensor, opt=self.opt)
self.criterionFeat = torch.nn.L1Loss()
if not opt.no_vgg_loss:
self.criterionVGG = networks.VGGLoss(self.opt.gpu_ids)
if opt.use_vae:
self.KLDLoss = networks.KLDLoss()
# Entry point for all calls involving forward pass
# of deep networks. We used this approach since DataParallel module
# can't parallelize custom functions, we branch to different
# routines based on |mode|.
def forward(self, data, mode,iters,progress,epochs,images_iters):
input_semantics, real_image = self.preprocess_input(data)
# print("input_semantics: ", input_semantics)
# print("real_image: ", real_image.shape)
if mode == 'generator':
g_loss, generated = self.compute_generator_loss(input_semantics, real_image,iters=iters,progress=progress,epoch_model=epochs,images_iter=images_iters)
return g_loss, generated
elif mode == 'discriminator':
d_loss, d2_loss = self.compute_discriminator_loss(input_semantics, real_image,iters=iters)
return d_loss, d2_loss
elif mode == 'encode_only':
z, mu, logvar = self.encode_z(real_image)
return mu, logvar
elif mode == 'inference':
contents = F.interpolate(real_image, size=(224, 224), mode='bilinear', align_corners=False)
styles = F.interpolate(input_semantics, size=(224, 224), mode='bilinear', align_corners=False)
model = self.generate_fake(contents, styles, iters, mode="gen",progress=None)
with torch.no_grad():
self.opt.alpha = self.opt.alpha.to(self.device)
_, _, _, _, output = model(contents, styles)
print("OUTPUT",output.shape)
upsample_layer = nn.Sequential(nn.Upsample(scale_factor=8 / 7, mode='bilinear', align_corners=False))
fake_image = upsample_layer(output)
return fake_image
else:
raise ValueError("|mode| is invalid")
def create_optimizers(self, opt):
G_params = list(self.netG.parameters())
if opt.use_vae:
G_params += list(self.netE.parameters())
if opt.isTrain:
D_params = list(self.netD.parameters())
D2_params = list(self.netD2.parameters())
if opt.no_TTUR:
beta1, beta2 = opt.beta1, opt.beta2
G_lr, D_lr = opt.lr, opt.lr
else:
beta1, beta2 = 0, 0.9
G_lr, D_lr = opt.lr / 2, opt.lr * 2
optimizer_G = torch.optim.Adam(G_params, lr=G_lr, betas=(beta1, beta2))
optimizer_D = torch.optim.Adam(D_params, lr=D_lr, betas=(beta1, beta2))
optimizer_D2 = torch.optim.Adam(D2_params, lr=D_lr, betas=(beta1, beta2))
return optimizer_G, optimizer_D, optimizer_D2
def save(self, epoch):
util.save_network(self.netG, 'G', epoch, self.opt)
util.save_network(self.netD, 'D', epoch, self.opt)
util.save_network(self.netD2, 'D2', epoch, self.opt)
if self.opt.use_vae:
util.save_network(self.netE, 'E', epoch, self.opt)
############################################################################
# Private helper methods
############################################################################
def initialize_networks(self, opt):
netG = networks.define_G(opt)
netD = networks.define_D(opt) if opt.isTrain else None
netD2 = networks.define_D(opt) if opt.isTrain else None
netE = networks.define_E(opt) if opt.use_vae else None
if not opt.isTrain or opt.continue_train:
netG = util.load_network(netG, 'G', opt.which_epoch, opt)
if opt.isTrain:
netD = util.load_network(netD, 'D', opt.which_epoch, opt)
netD2 = util.load_network(netD2, 'D2', opt.which_epoch, opt)
if opt.use_vae:
netE = util.load_network(netE, 'E', opt.which_epoch, opt)
return netG, netD, netD2, netE
# preprocess the input, such as moving the tensors to GPUs
# and transforming the label map to one-hot encoding (for SIS)
# |data|: dictionary of the input data
def preprocess_input(self, data):
# move to GPU and change data types
if self.opt.task == 'SIS':
data['label'] = data['label'].long()
if self.use_gpu():
data['label'] = data['label'].cuda()
if 'instance' in data:
try:
data['instance'] = data['instance'].cuda()
except:
data['instance'] = data['instance']
else:
data['instance'] = None # or handle as appropriate for your model
data['image'] = data['image'].cuda()
# create one-hot label map for SIS
if self.opt.task == 'SIS':
label_map = data['label']
bs, _, h, w = label_map.size()
nc = self.opt.label_nc + 1 if self.opt.contain_dontcare_label \
else self.opt.label_nc
input_label = self.FloatTensor(bs, nc, h, w).zero_()
input_semantics = input_label.scatter_(1, label_map, 1.0)
# concatenate instance map if it exists
if not self.opt.no_instance:
inst_map = data['instance']
instance_edge_map = self.get_edges(inst_map)
input_semantics = torch.cat((input_semantics, instance_edge_map), dim=1)
else:
input_semantics = data['label']
return input_semantics, data['image']
def compute_generator_loss(self, content, style,iters,progress,epoch_model,images_iter):
G_losses = {}
fake_image, loss_StyTr2 = self.generate_fake(content, style,iters=iters,progress=progress,epochs=epoch_model,image_iters=images_iter)
pred_fake, pred_real_c = self.discriminate(fake_image, content)
# # computer loss using Discriminator 1
G_losses['GAN'] = self.criterionGAN(pred_fake, True, for_discriminator=False) * (1 - self.opt.alpha)
# computer loss using Discriminator 2
pred_fake, pred_real = self.discriminate(fake_image, style, type='D2')
G_losses['GAN'] = self.criterionGAN(pred_fake, True, for_discriminator=False) * self.opt.alpha
if self.opt.task == 'SIS':
pred_fake, pred_real = self.discriminate(fake_image, style, content)
else:
pred_fake, pred_real = self.discriminate(fake_image, style)
G_losses['GAN'] = self.criterionGAN(pred_fake, True,for_discriminator=False)
gan_loss_value = G_losses['GAN'].item()
new_GAN_loss = gan_loss_value + loss_StyTr2
G_losses['GAN'] = torch.tensor([new_GAN_loss], device=self.device)
if not self.opt.no_ganFeat_loss:
num_D = len(pred_fake)
GAN_Feat_loss = torch.tensor(0.0, dtype=torch.float32, device=self.device)
for i in range(num_D): # for each discriminator
# last output is the final prediction, so we exclude it
num_intermediate_outputs = len(pred_fake[i]) - 1
for j in range(num_intermediate_outputs): # for each layer output
unweighted_loss = self.criterionFeat(
pred_fake[i][j], pred_real[i][j].detach())
GAN_Feat_loss += unweighted_loss * self.opt.lambda_feat / num_D
G_losses['GAN_Feat'] = GAN_Feat_loss
if not self.opt.no_vgg_loss:
target = style if self.opt.task == 'SIS' else content
G_losses['VGG'] = self.criterionVGG(fake_image, target) * self.opt.lambda_vgg
return G_losses, fake_image
def compute_discriminator_loss(self, content, style,iters):
D_losses = {}
D2_losses = {}
contents = F.interpolate(content, size=(224, 224), mode='bilinear', align_corners=False)
styles = F.interpolate(style, size=(224, 224), mode='bilinear', align_corners=False)
model=self.generate_fake(contents,styles,iters,mode="gen",progress=None)
with torch.no_grad():
_, _, _, _, output=model(contents,styles)
upsample_layer = nn.Sequential(nn.Upsample(scale_factor=8 / 7, mode='bilinear', align_corners=False))
fake_image = upsample_layer(output)
# For Discriminator 1 between fake and content
pred_fake, pred_real = self.discriminate(fake_image, content)
D_losses['D_Fake'] = self.criterionGAN(pred_fake, False, for_discriminator=True) * (1 - self.opt.alpha)
D_losses['D_real'] = self.criterionGAN(pred_real, True, for_discriminator=True)
pred_fake, pred_real = self.discriminate(fake_image, style, type='D2')
D2_losses['D_Fake'] = self.criterionGAN(pred_fake, False, for_discriminator=True) * self.opt.alpha
D2_losses['D_real'] = self.criterionGAN(pred_real, True, for_discriminator=True)
return D_losses, D2_losses
def encode_z(self, real_image):
mu, logvar = self.netE(real_image)
z = self.reparameterize(mu, logvar)
return z, mu, logvar
def custom_epoch(self, curr_index_image):
if curr_index_image <= 5:
return self.max_iter
elif 5 < curr_index_image < 15:
return int(self.max_iter * 0.6)
elif 15 <= curr_index_image < 25:
return int(self.max_iter * 0.55)
elif 25 <= curr_index_image < 40:
return int(self.max_iter * 0.5)
elif 40 <= curr_index_image < 60:
return int(self.max_iter * 0.375)
elif 60 <= curr_index_image < 80:
return int(self.max_iter * 0.25)
elif 80 <= curr_index_image < 120:
return int(self.max_iter * 0.1)
else:
return int(self.max_iter * 0.05)
def save_generator_weights_by_iters(self,net, label, epoch,iters):
util.save_generator_by_iter(net,label,epoch,iters,self.opt)
def generate_fake(self, input_semantics, real_image, iters,mode="train",progress=None,epochs=None,image_iters=None):
### FOR GENERATE-FAKE IMAGES
if mode == "train":
models_Generator = self.netG()
optimizer = torch.optim.Adam([
{'params': models_Generator.SEencoder.parameters()},
{'params': models_Generator.decoder.parameters()},
{'params': models_Generator.transModule.parameters()},
], lr=self.lrG_decay)
scheduler = CosineAnnealingWarmUpLR(optimizer, warmup_step=self.max_iter // 4, max_step=self.max_iter,
min_lr=0)
contents = real_image.to(self.device)
style = input_semantics.to(self.device)
content_images = F.interpolate(contents, size=(224, 224), mode='bilinear', align_corners=False)
style_images = F.interpolate(style, size=(224, 224), mode='bilinear', align_corners=False)
total_loss = 0
# Mở file ghi log
log_file = open("train_log.txt", "a")
with progress:
task = progress.add_task(
f"[green]Iters on one Images [{iters}|{self.custom_epoch(iters)}]- Total_Loss: {total_loss}",
total=self.custom_epoch(iters)
)
total_i = 0
step_wb=0
for i in range(self.custom_epoch(curr_index_image=iters)):
# Cộng dồn giá trị i vào tổng
if (i-self.custom_epoch(curr_index_image=iters)):
total_i=i
step_wb+=1
if (step_wb>=1):
total_i=total_i+1
loss_c, loss_s, loss_id_1, loss_id_2, out = models_Generator(content_images, style_images)
self.out = out
loss_all = (self.content_weight * loss_c + self.style_weight * loss_s +
self.id1_weight * loss_id_1 + self.id2_weight * loss_id_2)
# Tính tổng loss
total_loss = round(float(loss_all.sum().cpu().detach().numpy()), 3)
print("Loss_model", loss_all.sum().cpu().detach().numpy(), "==>Content_Loss",
loss_c.sum().cpu().detach().numpy(),
"==>Style_Loss", loss_s.sum().cpu().detach().numpy(), "==>ID_1_Loss",
loss_id_1.sum().cpu().detach().numpy(),
"==>ID_2_Loss", loss_id_2.sum().cpu().detach().numpy())
#### WANDB ONE IMAGES
# wblogging.upload_all_loss_on_one_images(loss_value=total_loss, iters_all_images=image_iters,
# iters_one_imgs=total_i, epoch=epochs)
# wblogging.upload_l1_loss_on_one_images(
# loss_value=round(float(loss_id_1.sum().cpu().detach().numpy()), 3),
# iters_all_images=image_iters, iters_one_imgs=total_i, epoch=epochs)
# wblogging.upload_l2_loss_on_one_images(
# loss_value=round(float(loss_id_2.sum().cpu().detach().numpy()), 3),
# iters_all_images=image_iters, iters_one_imgs=total_i, epoch=epochs)
# wblogging.upload_content_loss_on_one_images(
# loss_value=round(float(loss_c.sum().cpu().detach().numpy()), 3),
# iters_all_images=image_iters, iters_one_imgs=total_i, epoch=epochs)
# wblogging.upload_style_loss_on_one_images(
# loss_value=round(float(loss_s.sum().cpu().detach().numpy()), 3),
# iters_all_images=image_iters, iters_one_imgs=total_i, epoch=epochs)
current_lr = optimizer.param_groups[0]['lr']
wblogging.upload_lr(current_lr, epoch=epochs)
### END LOGGING ON ONE IMAGES
# Lấy thời gian hiện tại với format [ss/mm/hh/dd/mm/yyyy]
current_time = datetime.datetime.now().strftime("[%S-%M-%H-%d-%m-%Y]")
# Ghi log vào file
log_file.write(
f"{current_time} Iter: {i}, Total Loss: {total_loss}, Content Loss: {loss_c.sum().cpu().detach().numpy()}, "
f"Style Loss: {loss_s.sum().cpu().detach().numpy()}, ID1 Loss: {loss_id_1.sum().cpu().detach().numpy()}, "
f"ID2 Loss: {loss_id_2.sum().cpu().detach().numpy()}\n")
progress.update(task, advance=1,
description=f"[green]Iters on one Images [{i + 1}|{self.custom_epoch(iters)}]- Total_Loss: {total_loss}")
# Update parameters
optimizer.zero_grad()
loss_all.backward()
optimizer.step()
scheduler.step()
# SAVE Weight by Iters
if ((i + 1) - (self.custom_epoch(curr_index_image=iters))) == 0:
self.save_generator_weights_by_iters(self.netG, 'G', epoch=self.epoch, iters=i + 1)
if i % 100 == 0:
output_name = f'{self.save_dir}/test/{str(i)}.jpg'
out = torch.cat((content_images, out), 0)
out = torch.cat((style_images, out), 0)
save_image(out, output_name)
# Đóng file log sau khi kết thúc
log_file.close()
# #### WANDB ALL IMAGES
# wblogging.upload_loss_all(loss_all=total_loss, iters=image_iters, epoch=epochs)
# wblogging.upload_l1_loss_all(loss_values=round(float(loss_id_1.sum().cpu().detach().numpy()), 3),
# iters_images=image_iters, epoch=epochs)
# wblogging.upload_l2_loss(loss_values=round(float(loss_id_2.sum().cpu().detach().numpy()), 3),
# iters=image_iters, epoch=epochs)
# wblogging.upload_content_loss(loss_c=round(float(loss_c.sum().cpu().detach().numpy()), 3),
# iters=image_iters, epoch=epochs)
# wblogging.upload_style_loss(loss_s=round(float(loss_id_2.sum().cpu().detach().numpy()), 3),
# iters=image_iters, epoch=epochs)
### END WANDB ON ALL IMAGES
upsample_layer = nn.Sequential(nn.Upsample(scale_factor=8 / 7, mode='bilinear', align_corners=False))
out = upsample_layer(self.out)
return out, self.loss
# FOR GENERATOR IMAGES - VAL
if(mode=="gen"):
models_Generator = self.netG()
models_Generator.eval()
models_Generator.to(self.device)
return models_Generator
# Given fake and real image, return the prediction of discriminator
# for each fake and real image. The condition is used in SIS.
def discriminate(self, fake_image, real_image, condition=None,type='D1'):
if self.opt.task == 'SIS':
assert condition is not None
fake_concat = torch.cat([condition, fake_image], dim=1)
real_concat = torch.cat([condition, real_image], dim=1)
else:
assert condition is None
fake_concat = fake_image
real_concat = real_image
# In Batch Normalization, the fake and real images are
# recommended to be in the same batch to avoid disparate
# statistics in fake and real images.
# So both fake and real images are fed to D all at once.
fake_and_real = torch.cat([fake_concat, real_concat], dim=0)
if type == 'D1':
discriminator_out = self.netD(fake_and_real)
else:
discriminator_out = self.netD2(fake_and_real)
pred_fake, pred_real = self.divide_pred(discriminator_out)
return pred_fake, pred_real
# Take the prediction of fake and real images from the combined batch
def divide_pred(self, pred):
# the prediction contains the intermediate outputs of multi-scale GAN,
# so it's usually a list
if type(pred) == list:
fake = []
real = []
for p in pred:
fake.append([tensor[:tensor.size(0) // 2] for tensor in p])
real.append([tensor[tensor.size(0) // 2:] for tensor in p])
else:
fake = pred[:pred.size(0) // 2]
real = pred[pred.size(0) // 2:]
return fake, real
def get_edges(self, t):
edge = self.ByteTensor(t.size()).zero_()
edge[:, :, :, 1:] = edge[:, :, :, 1:] | (t[:, :, :, 1:] != t[:, :, :, :-1])
edge[:, :, :, :-1] = edge[:, :, :, :-1] | (t[:, :, :, 1:] != t[:, :, :, :-1])
edge[:, :, 1:, :] = edge[:, :, 1:, :] | (t[:, :, 1:, :] != t[:, :, :-1, :])
edge[:, :, :-1, :] = edge[:, :, :-1, :] | (t[:, :, 1:, :] != t[:, :, :-1, :])
return edge.float()
def reparameterize(self, mu, logvar):
std = torch.exp(0.5 * logvar)
eps = torch.randn_like(std)
return eps.mul(std) + mu
def use_gpu(self):
return len(self.opt.gpu_ids) > 0
|