admin/Dehaze
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

整合去雾网页工具

Browse Source
This commit is contained in:
admin
2026-06-10 17:42:11 +08:00
commit 6db15ebc3f
101 changed files with 10167 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

67
RefineDNet/models/__init__.py Normal file
View File

@@ -0,0 +1,67 @@
"""This package contains modules related to objective functions, optimizations, and network architectures.
To add a custom model class called 'dummy', you need to add a file called 'dummy_model.py' and define a subclass DummyModel inherited from BaseModel.
You need to implement the following five functions:
-- <__init__>: initialize the class; first call BaseModel.__init__(self, opt).
-- <set_input>: unpack data from dataset and apply preprocessing.
-- <forward>: produce intermediate results.
-- <optimize_parameters>: calculate loss, gradients, and update network weights.
-- <modify_commandline_options>: (optionally) add model-specific options and set default options.
In the function <__init__>, you need to define four lists:
-- self.loss_names (str list): specify the training losses that you want to plot and save.
-- self.model_names (str list): define networks used in our training.
-- self.visual_names (str list): specify the images that you want to display and save.
-- self.optimizers (optimizer list): define and initialize optimizers. You can define one optimizer for each network. If two networks are updated at the same time, you can use itertools.chain to group them. See cycle_gan_model.py for an usage.
Now you can use the model class by specifying flag '--model dummy'.
See our template model class 'template_model.py' for more details.
"""
import importlib
from models.base_model import BaseModel
def find_model_using_name(model_name):
"""Import the module "models/[model_name]_model.py".
In the file, the class called DatasetNameModel() will
be instantiated. It has to be a subclass of BaseModel,
and it is case-insensitive.
"""
model_filename = "models." + model_name + "_model"
modellib = importlib.import_module(model_filename)
model = None
target_model_name = model_name.replace('_', '') + 'model'
for name, cls in modellib.__dict__.items():
if name.lower() == target_model_name.lower() \
and issubclass(cls, BaseModel):
model = cls
if model is None:
print("In %s.py, there should be a subclass of BaseModel with class name that matches %s in lowercase." % (model_filename, target_model_name))
exit(0)
return model
def get_option_setter(model_name):
"""Return the static method <modify_commandline_options> of the model class."""
model_class = find_model_using_name(model_name)
return model_class.modify_commandline_options
def create_model(opt):
"""Create a model given the option.
This function warps the class CustomDatasetDataLoader.
This is the main interface between this package and 'train.py'/'test.py'
Example:
>>> from models import create_model
>>> model = create_model(opt)
"""
model = find_model_using_name(opt.model)
instance = model(opt)
print("model [%s] was created" % type(instance).__name__)
return instance

229
RefineDNet/models/base_model.py Normal file
View File

@@ -0,0 +1,229 @@
import os
import torch
from collections import OrderedDict
from abc import ABC, abstractmethod
from . import networks
class BaseModel(ABC):
"""This class is an abstract base class (ABC) for models.
To create a subclass, you need to implement the following five functions:
-- <__init__>: initialize the class; first call BaseModel.__init__(self, opt).
-- <set_input>: unpack data from dataset and apply preprocessing.
-- <forward>: produce intermediate results.
-- <optimize_parameters>: calculate losses, gradients, and update network weights.
-- <modify_commandline_options>: (optionally) add model-specific options and set default options.
"""
def __init__(self, opt):
"""Initialize the BaseModel class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
When creating your custom class, you need to implement your own initialization.
In this fucntion, you should first call <BaseModel.__init__(self, opt)>
Then, you need to define four lists:
-- self.loss_names (str list): specify the training losses that you want to plot and save.
-- self.model_names (str list): specify the images that you want to display and save.
-- self.visual_names (str list): define networks used in our training.
-- self.optimizers (optimizer list): define and initialize optimizers. You can define one optimizer for each network. If two networks are updated at the same time, you can use itertools.chain to group them. See cycle_gan_model.py for an example.
"""
self.opt = opt
self.gpu_ids = opt.gpu_ids
self.isTrain = opt.isTrain
self.device = torch.device('cuda:{}'.format(self.gpu_ids[0])) if self.gpu_ids else torch.device('cpu') # get device name: CPU or GPU
self.save_dir = os.path.join(opt.checkpoints_dir, opt.name) # save all the checkpoints to save_dir
if opt.preprocess != 'scale_width': # with [scale_width], input images might have different sizes, which hurts the performance of cudnn.benchmark.
torch.backends.cudnn.benchmark = True
self.loss_names = []
self.model_names = []
self.visual_names = []
self.optimizers = []
self.image_paths = []
self.metric = 0 # used for learning rate policy 'plateau'
@staticmethod
def modify_commandline_options(parser, is_train):
"""Add new model-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
"""
return parser
@abstractmethod
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input (dict): includes the data itself and its metadata information.
"""
pass
@abstractmethod
def forward(self):
"""Run forward pass; called by both functions <optimize_parameters> and <test>."""
pass
@abstractmethod
def optimize_parameters(self):
"""Calculate losses, gradients, and update network weights; called in every training iteration"""
pass
def setup(self, opt):
"""Load and print networks; create schedulers
Parameters:
opt (Option class) -- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
if self.isTrain:
self.schedulers = [networks.get_scheduler(optimizer, opt) for optimizer in self.optimizers]
if not self.isTrain or opt.continue_train:
load_suffix = 'iter_%d' % opt.load_iter if opt.load_iter > 0 else opt.epoch
self.load_networks(load_suffix)
self.print_networks(opt.verbose)
def eval(self):
"""Make models eval mode during test time"""
for name in self.model_names:
if isinstance(name, str):
net = getattr(self, 'net' + name)
net.eval()
def test(self):
"""Forward function used in test time.
This function wraps <forward> function in no_grad() so we don't save intermediate steps for backprop
It also calls <compute_visuals> to produce additional visualization results
"""
with torch.no_grad():
self.forward()
self.compute_visuals()
def compute_visuals(self):
"""Calculate additional output images for visdom and HTML visualization"""
pass
def get_image_paths(self):
""" Return image paths that are used to load current data"""
return self.image_paths
def update_learning_rate(self):
"""Update learning rates for all the networks; called at the end of every epoch"""
for scheduler in self.schedulers:
if self.opt.lr_policy == 'plateau':
scheduler.step(self.metric)
else:
scheduler.step()
lr = self.optimizers[0].param_groups[0]['lr']
print('learning rate = %.7f' % lr)
def get_current_visuals(self):
"""Return visualization images. train.py will display these images with visdom, and save the images to a HTML"""
visual_ret = OrderedDict()
for name in self.visual_names:
if isinstance(name, str):
visual_ret[name] = getattr(self, name)
return visual_ret
def get_current_losses(self):
"""Return traning losses / errors. train.py will print out these errors on console, and save them to a file"""
errors_ret = OrderedDict()
for name in self.loss_names:
if isinstance(name, str):
errors_ret[name] = float(getattr(self, 'loss_' + name)) # float(...) works for both scalar tensor and float number
return errors_ret
def save_networks(self, epoch):
"""Save all the networks to the disk.
Parameters:
epoch (int) -- current epoch; used in the file name '%s_net_%s.pth' % (epoch, name)
"""
for name in self.model_names:
if isinstance(name, str):
save_filename = '%s_net_%s.pth' % (epoch, name)
save_path = os.path.join(self.save_dir, save_filename)
net = getattr(self, 'net' + name)
if len(self.gpu_ids) > 0 and torch.cuda.is_available():
torch.save(net.module.cpu().state_dict(), save_path)
net.cuda(self.gpu_ids[0])
else:
torch.save(net.cpu().state_dict(), save_path)
def __patch_instance_norm_state_dict(self, state_dict, module, keys, i=0):
"""Fix InstanceNorm checkpoints incompatibility (prior to 0.4)"""
key = keys[i]
if i + 1 == len(keys): # at the end, pointing to a parameter/buffer
if module.__class__.__name__.startswith('InstanceNorm') and \
(key == 'running_mean' or key == 'running_var'):
if getattr(module, key) is None:
state_dict.pop('.'.join(keys))
if module.__class__.__name__.startswith('InstanceNorm') and \
(key == 'num_batches_tracked'):
state_dict.pop('.'.join(keys))
else:
self.__patch_instance_norm_state_dict(state_dict, getattr(module, key), keys, i + 1)
def load_networks(self, epoch):
"""Load all the networks from the disk.
Parameters:
epoch (int) -- current epoch; used in the file name '%s_net_%s.pth' % (epoch, name)
"""
for name in self.model_names:
if isinstance(name, str):
load_filename = '%s_net_%s.pth' % (epoch, name)
load_path = os.path.join(self.save_dir, load_filename)
net = getattr(self, 'net' + name)
if isinstance(net, torch.nn.DataParallel):
net = net.module
print('loading the model from %s' % load_path)
# if you are using PyTorch newer than 0.4 (e.g., built from
# GitHub source), you can remove str() on self.device
state_dict = torch.load(load_path, map_location=str(self.device))
if hasattr(state_dict, '_metadata'):
del state_dict._metadata
# patch InstanceNorm checkpoints prior to 0.4
for key in list(state_dict.keys()): # need to copy keys here because we mutate in loop
self.__patch_instance_norm_state_dict(state_dict, net, key.split('.'))
net.load_state_dict(state_dict)
def print_networks(self, verbose):
"""Print the total number of parameters in the network and (if verbose) network architecture
Parameters:
verbose (bool) -- if verbose: print the network architecture
"""
print('---------- Networks initialized -------------')
for name in self.model_names:
if isinstance(name, str):
net = getattr(self, 'net' + name)
num_params = 0
for param in net.parameters():
num_params += param.numel()
if verbose:
print(net)
print('[Network %s] Total number of parameters : %.3f M' % (name, num_params / 1e6))
print('-----------------------------------------------')
def set_requires_grad(self, nets, requires_grad=False):
"""Set requies_grad=Fasle for all the networks to avoid unnecessary computations
Parameters:
nets (network list) -- a list of networks
requires_grad (bool) -- whether the networks require gradients or not
"""
if not isinstance(nets, list):
nets = [nets]
for net in nets:
if net is not None:
for param in net.parameters():
param.requires_grad = requires_grad

221
RefineDNet/models/basic_dehaze_model.py Normal file
View File

@@ -0,0 +1,221 @@
import torch
import itertools
from util.image_pool import ImagePool
from .base_model import BaseModel
from . import networks
import torch.nn.functional as F
from util import util
class BasicDehazeModel(BaseModel):
"""
This class implements the CycleGAN model, for learning image-to-image translation without paired data.
The model training requires '--dataset_mode unaligned' dataset.
By default, it uses a '--netG resnet_9blocks' ResNet generator,
a '--netD basic' discriminator (PatchGAN introduced by pix2pix),
and a least-square GANs objective ('--gan_mode lsgan').
CycleGAN paper: https://arxiv.org/pdf/1703.10593.pdf
"""
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new dataset-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
For CycleGAN, in addition to GAN losses, we introduce lambda_A, lambda_B, and lambda_identity for the following losses.
A (source domain), B (target domain).
Generators: G_A: A -> B; G_B: B -> A.
Discriminators: D_A: G_A(A) vs. B; D_B: G_B(B) vs. A.
Forward cycle loss: lambda_A * ||G_B(G_A(A)) - A|| (Eqn. (2) in the paper)
Backward cycle loss: lambda_B * ||G_A(G_B(B)) - B|| (Eqn. (2) in the paper)
Identity loss (optional): lambda_identity * (||G_A(B) - B|| * lambda_B + ||G_B(A) - A|| * lambda_A) (Sec 5.2 "Photo generation from paintings" in the paper)
Dropout is not used in the original CycleGAN paper.
"""
parser.set_defaults(no_dropout=True) # default CycleGAN did not use dropout
if is_train:
parser.add_argument('--lambda_haze', type=float, default=0.1, help='weight for D_haze')
parser.add_argument('--lambda_clear', type=float, default=0.1, help='weight for D_clear')
parser.add_argument('--lambda_tv', type=float, default=1, help='weight for D_clear')
parser.add_argument('--lambda_identity', type=float, default=0.5, help='use identity mapping. Setting lambda_identity other than 0 has an effect of scaling the weight of the identity mapping loss. For example, if the weight of the identity loss should be 10 times smaller than the weight of the reconstruction loss, please set lambda_identity = 0.1')
parser.add_argument('--netR_T', type=str, default='unet_trans_256', help='specify generator architecture')
parser.add_argument('--netR_J', type=str, default='haze_refine_2', help='specify generator architecture')
return parser
def __init__(self, opt):
"""Initialize the CycleGAN class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
BaseModel.__init__(self, opt)
# specify the training losses you want to print out. The training/test scripts will call <BaseModel.get_current_losses>
self.loss_names = ['D_haze', 'G_rec_I', 'D_clear', 'G_ref_J', 'rec_I', 'rec_J', 'TV_T', 'idt_J']
# specify the images you want to save/display. The training/test scripts will call <BaseModel.get_current_visuals>
self.visual_names = ['real_I', 'est_J', 'rec_I', 'rec_J',
'est_T_vis', 'out_T_vis', 'real_J']
# specify the models you want to save to the disk. The training/test scripts will call <BaseModel.save_networks> and <BaseModel.load_networks>.
if self.isTrain:
self.model_names = ['Est_T', 'Est_J', 'D_haze', 'D_clear']
else: # during test time, only load Gs
self.model_names = ['Est_T', 'Est_J']
# define networks (both Generators and discriminators)
self.netEst_T = networks.define_G(opt.input_nc, 1, opt.ngf, opt.netR_T, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
self.netEst_J = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netR_J, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain: # define discriminators
self.netD_haze = networks.define_D(opt.input_nc, opt.ndf, opt.netD,
opt.n_layers_D, opt.norm, opt.init_type, opt.init_gain, self.gpu_ids)
self.netD_clear = networks.define_D(opt.output_nc, opt.ndf, opt.netD,
opt.n_layers_D, opt.norm, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain:
if opt.lambda_identity > 0.0: # only works when input and output images have the same number of channels
assert(opt.input_nc == opt.output_nc)
self.fake_I_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images
self.fake_J_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images
# # define loss functions
self.criterionGAN = networks.GANLoss(opt.gan_mode).to(self.device) # define GAN loss.
self.criterionRec = torch.nn.L1Loss()
self.criterionIdt = torch.nn.L1Loss()
self.criterionTV = networks.TVLoss()
# initialize optimizers; schedulers will be automatically created by function <BaseModel.setup>.
self.optimizer_G = torch.optim.Adam(itertools.chain(self.netEst_T.parameters(), self.netEst_J.parameters()), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizer_D = torch.optim.Adam(itertools.chain(self.netD_haze.parameters(), self.netD_clear.parameters()), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizers.append(self.optimizer_G)
self.optimizers.append(self.optimizer_D)
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input (dict): include the data itself and its metadata information.
The option 'direction' can be used to swap domain A and domain B.
"""
self.real_I = input['haze'].to(self.device) # [-1, 1]
self.real_J = input['clear'].to(self.device) # [-1, 1]
self.image_paths = input['paths']
def forward(self):
"""Run forward pass; called by both functions <optimize_parameters> and <test>."""
# output scale [0,1]
self.est_T, self.out_T = self.netEst_T(self.real_I)
self.est_J = self.netEst_J(self.real_I)
# reconstruct haze image
est_T_map = self.est_T.repeat(1,3,1,1)
self.rec_I = util.synthesize_fog(self.est_J, est_T_map)
self.rec_J = util.reverse_fog(self.real_I, est_T_map)
def test(self):
"""Forward function used in test time.
This function wraps <forward> function in no_grad() so we don't save intermediate steps for backprop
It also calls <compute_visuals> to produce additional visualization results
"""
with torch.no_grad():
self.forward()
self.compute_visuals()
self.refine_J = (self.rec_J + self.est_J)/2
self.visual_names.append('refine_J')
def compute_visuals(self):
"""Calculate additional output images for visdom and HTML visualization"""
# rescale to [-1,1] for visdom
self.est_T_vis = (self.est_T - 0.5)/0.5
self.out_T_vis = (self.out_T - 0.5)/0.5
# self.map_A_vis = (self.map_A - 0.5)/0.5
def backward_D_basic(self, netD, real, fake):
"""Calculate GAN loss for the discriminator
Parameters:
netD (network) -- the discriminator D
real (tensor array) -- real images
fake (tensor array) -- images generated by a generator
Return the discriminator loss.
We also call loss_D.backward() to calculate the gradients.
"""
# Real
pred_real = netD(real)
loss_D_real = self.criterionGAN(pred_real, True)
# Fake
pred_fake = netD(fake.detach())
loss_D_fake = self.criterionGAN(pred_fake, False)
# Combined loss and calculate gradients
loss_D = (loss_D_real + loss_D_fake) * 0.5
loss_D.backward()
return loss_D
def backward_D_haze(self):
fake_I = self.fake_I_pool.query(self.rec_I)
self.loss_D_haze = self.backward_D_basic(self.netD_haze, self.real_I, fake_I)
def backward_D_clear(self):
fake_J = self.fake_J_pool.query(self.est_J)
self.loss_D_clear = self.backward_D_basic(self.netD_clear, self.real_J, fake_J)
def backward_G(self):
lambda_idt = self.opt.lambda_identity
lambda_tv = self.opt.lambda_tv
lambda_haze = self.opt.lambda_haze
lambda_clear = self.opt.lambda_clear
# TV loss
if lambda_tv > 0.0:
self.loss_TV_T = self.criterionTV(self.out_T)*lambda_tv
else:
self.loss_TV_T = 0
# Identity loss
if lambda_idt > 0.0:
self.loss_idt_J = self.criterionIdt(self.netEst_J(self.real_J), self.real_J)*lambda_idt
else:
self.loss_idt_J = 0
# Generator losses for rec_I and est_J
self.loss_G_rec_I = self.criterionGAN(self.netD_haze(self.rec_I), True)*lambda_haze
self.loss_G_ref_J = self.criterionGAN(self.netD_clear(self.est_J), True)*lambda_clear #+ \
# self.criterionGAN(self.netD_clear(self.rec_J), True)*lambda_clear
# Reconstrcut loss
self.loss_rec_I = self.criterionRec(self.rec_I, self.real_I)
# only compute, not back propagate
self.loss_rec_J = self.criterionRec(self.rec_J, self.est_J)
self.loss_G = self.loss_G_rec_I + self.loss_G_ref_J + self.loss_rec_I + self.loss_idt_J + self.loss_TV_T
self.loss_G.backward()
def optimize_parameters(self):
"""Calculate losses, gradients, and update network weights; called in every training iteration"""
# forward
self.forward() # compute fake images and reconstruction images.
# G_A and G_B
self.set_requires_grad([self.netD_haze, self.netD_clear], False) # Ds require no gradients when optimizing Gs
self.optimizer_G.zero_grad() # set G_A and G_B's gradients to zero
self.backward_G() # calculate gradients for G_A and G_B
self.optimizer_G.step() # update G_A and G_B's weights
# D_A and D_B
self.set_requires_grad([self.netD_haze, self.netD_clear], True)
self.optimizer_D.zero_grad() # set D_A and D_B's gradients to zero
self.backward_D_haze() # calculate gradients for D_A
self.backward_D_clear() # calculate graidents for D_B
self.optimizer_D.step() # update D_A and D_B's weights

68
RefineDNet/models/colorization_model.py Normal file
View File

@@ -0,0 +1,68 @@
from .pix2pix_model import Pix2PixModel
import torch
from skimage import color # used for lab2rgb
import numpy as np
class ColorizationModel(Pix2PixModel):
"""This is a subclass of Pix2PixModel for image colorization (black & white image -> colorful images).
The model training requires '-dataset_model colorization' dataset.
It trains a pix2pix model, mapping from L channel to ab channels in Lab color space.
By default, the colorization dataset will automatically set '--input_nc 1' and '--output_nc 2'.
"""
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new dataset-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
By default, we use 'colorization' dataset for this model.
See the original pix2pix paper (https://arxiv.org/pdf/1611.07004.pdf) and colorization results (Figure 9 in the paper)
"""
Pix2PixModel.modify_commandline_options(parser, is_train)
parser.set_defaults(dataset_mode='colorization')
return parser
def __init__(self, opt):
"""Initialize the class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
For visualization, we set 'visual_names' as 'real_A' (input real image),
'real_B_rgb' (ground truth RGB image), and 'fake_B_rgb' (predicted RGB image)
We convert the Lab image 'real_B' (inherited from Pix2pixModel) to a RGB image 'real_B_rgb'.
we convert the Lab image 'fake_B' (inherited from Pix2pixModel) to a RGB image 'fake_B_rgb'.
"""
# reuse the pix2pix model
Pix2PixModel.__init__(self, opt)
# specify the images to be visualized.
self.visual_names = ['real_A', 'real_B_rgb', 'fake_B_rgb']
def lab2rgb(self, L, AB):
"""Convert an Lab tensor image to a RGB numpy output
Parameters:
L (1-channel tensor array): L channel images (range: [-1, 1], torch tensor array)
AB (2-channel tensor array): ab channel images (range: [-1, 1], torch tensor array)
Returns:
rgb (RGB numpy image): rgb output images (range: [0, 255], numpy array)
"""
AB2 = AB * 110.0
L2 = (L + 1.0) * 50.0
Lab = torch.cat([L2, AB2], dim=1)
Lab = Lab[0].data.cpu().float().numpy()
Lab = np.transpose(Lab.astype(np.float64), (1, 2, 0))
rgb = color.lab2rgb(Lab) * 255
return rgb
def compute_visuals(self):
"""Calculate additional output images for visdom and HTML visualization"""
self.real_B_rgb = self.lab2rgb(self.real_A, self.real_B)
self.fake_B_rgb = self.lab2rgb(self.real_A, self.fake_B)

211
RefineDNet/models/cycle_gan_model.py Normal file
View File

@@ -0,0 +1,211 @@
import torch
import itertools
from util.image_pool import ImagePool
from .base_model import BaseModel
from . import networks
class CycleGANModel(BaseModel):
"""
This class implements the CycleGAN model, for learning image-to-image translation without paired data.
The model training requires '--dataset_mode unaligned' dataset.
By default, it uses a '--netG resnet_9blocks' ResNet generator,
a '--netD basic' discriminator (PatchGAN introduced by pix2pix),
and a least-square GANs objective ('--gan_mode lsgan').
CycleGAN paper: https://arxiv.org/pdf/1703.10593.pdf
"""
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new dataset-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
For CycleGAN, in addition to GAN losses, we introduce lambda_A, lambda_B, and lambda_identity for the following losses.
A (source domain), B (target domain).
Generators: G_A: A -> B; G_B: B -> A.
Discriminators: D_A: G_A(A) vs. B; D_B: G_B(B) vs. A.
Forward cycle loss: lambda_A * ||G_B(G_A(A)) - A|| (Eqn. (2) in the paper)
Backward cycle loss: lambda_B * ||G_A(G_B(B)) - B|| (Eqn. (2) in the paper)
Identity loss (optional): lambda_identity * (||G_A(B) - B|| * lambda_B + ||G_B(A) - A|| * lambda_A) (Sec 5.2 "Photo generation from paintings" in the paper)
Dropout is not used in the original CycleGAN paper.
"""
parser.set_defaults(no_dropout=True) # default CycleGAN did not use dropout
if is_train:
parser.add_argument('--lambda_A', type=float, default=10.0, help='weight for cycle loss (A -> B -> A)')
parser.add_argument('--lambda_B', type=float, default=10.0, help='weight for cycle loss (B -> A -> B)')
parser.add_argument('--lambda_identity', type=float, default=0.5, help='use identity mapping. Setting lambda_identity other than 0 has an effect of scaling the weight of the identity mapping loss. For example, if the weight of the identity loss should be 10 times smaller than the weight of the reconstruction loss, please set lambda_identity = 0.1')
return parser
def __init__(self, opt):
"""Initialize the CycleGAN class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
BaseModel.__init__(self, opt)
# specify the training losses you want to print out. The training/test scripts will call <BaseModel.get_current_losses>
self.loss_names = ['D_A', 'G_A', 'cycle_A', 'idt_A', 'D_B', 'G_B', 'cycle_B', 'idt_B']
# specify the images you want to save/display. The training/test scripts will call <BaseModel.get_current_visuals>
visual_names_A = ['real_A', 'fake_B', 'rec_A']
visual_names_B = ['real_B', 'fake_A', 'rec_B']
if self.isTrain and self.opt.lambda_identity > 0.0: # if identity loss is used, we also visualize idt_B=G_A(B) ad idt_A=G_A(B)
visual_names_A.append('idt_B')
visual_names_B.append('idt_A')
self.visual_names = visual_names_A + visual_names_B # combine visualizations for A and B
# specify the models you want to save to the disk. The training/test scripts will call <BaseModel.save_networks> and <BaseModel.load_networks>.
if self.isTrain:
self.model_names = ['G_A', 'G_B', 'D_A', 'D_B']
else: # during test time, only load Gs
self.model_names = ['G_A', 'G_B']
# define networks (both Generators and discriminators)
# The naming is different from those used in the paper.
# Code (vs. paper): G_A (G), G_B (F), D_A (D_Y), D_B (D_X)
self.netG_A = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netG, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
self.netG_B = networks.define_G(opt.output_nc, opt.input_nc, opt.ngf, opt.netG, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain: # define discriminators
self.netD_A = networks.define_D(opt.output_nc, opt.ndf, opt.netD,
opt.n_layers_D, opt.norm, opt.init_type, opt.init_gain, self.gpu_ids)
self.netD_B = networks.define_D(opt.input_nc, opt.ndf, opt.netD,
opt.n_layers_D, opt.norm, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain:
if opt.lambda_identity > 0.0: # only works when input and output images have the same number of channels
assert(opt.input_nc == opt.output_nc)
self.fake_A_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images
self.fake_B_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images
# define loss functions
self.criterionGAN = networks.GANLoss(opt.gan_mode).to(self.device) # define GAN loss.
self.criterionCycle = torch.nn.L1Loss()
self.criterionIdt = torch.nn.L1Loss()
# initialize optimizers; schedulers will be automatically created by function <BaseModel.setup>.
self.optimizer_G = torch.optim.Adam(itertools.chain(self.netG_A.parameters(), self.netG_B.parameters()), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizer_D = torch.optim.Adam(itertools.chain(self.netD_A.parameters(), self.netD_B.parameters()), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizers.append(self.optimizer_G)
self.optimizers.append(self.optimizer_D)
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input (dict): include the data itself and its metadata information.
The option 'direction' can be used to swap domain A and domain B.
"""
if hasattr(self.opt, 'prior_name'):
self.real_A = input['haze'].to(self.device)
self.real_B = input['clear'].to(self.device)
self.image_paths = input['paths']
else:
AtoB = self.opt.direction == 'AtoB'
self.real_A = input['A' if AtoB else 'B'].to(self.device)
self.real_B = input['B' if AtoB else 'A'].to(self.device)
self.image_paths = input['A_paths' if AtoB else 'B_paths']
def forward(self):
"""Run forward pass; called by both functions <optimize_parameters> and <test>."""
self.fake_B = self.netG_A(self.real_A) # G_A(A)
self.rec_A = self.netG_B(self.fake_B) # G_B(G_A(A))
self.fake_A = self.netG_B(self.real_B) # G_B(B)
self.rec_B = self.netG_A(self.fake_A) # G_A(G_B(B))
def test(self):
"""Forward function used in test time.
This function wraps <forward> function in no_grad() so we don't save intermediate steps for backprop
It also calls <compute_visuals> to produce additional visualization results
"""
with torch.no_grad():
self.forward()
self.compute_visuals()
self.visual_names.append('refine_J')
self.refine_J = self.fake_B
def backward_D_basic(self, netD, real, fake):
"""Calculate GAN loss for the discriminator
Parameters:
netD (network) -- the discriminator D
real (tensor array) -- real images
fake (tensor array) -- images generated by a generator
Return the discriminator loss.
We also call loss_D.backward() to calculate the gradients.
"""
# Real
pred_real = netD(real)
loss_D_real = self.criterionGAN(pred_real, True)
# Fake
pred_fake = netD(fake.detach())
loss_D_fake = self.criterionGAN(pred_fake, False)
# Combined loss and calculate gradients
loss_D = (loss_D_real + loss_D_fake) * 0.5
loss_D.backward()
return loss_D
def backward_D_A(self):
"""Calculate GAN loss for discriminator D_A"""
fake_B = self.fake_B_pool.query(self.fake_B)
self.loss_D_A = self.backward_D_basic(self.netD_A, self.real_B, fake_B)
def backward_D_B(self):
"""Calculate GAN loss for discriminator D_B"""
fake_A = self.fake_A_pool.query(self.fake_A)
self.loss_D_B = self.backward_D_basic(self.netD_B, self.real_A, fake_A)
def backward_G(self):
"""Calculate the loss for generators G_A and G_B"""
lambda_idt = self.opt.lambda_identity
lambda_A = self.opt.lambda_A
lambda_B = self.opt.lambda_B
# Identity loss
if lambda_idt > 0:
# G_A should be identity if real_B is fed: ||G_A(B) - B||
self.idt_A = self.netG_A(self.real_B)
self.loss_idt_A = self.criterionIdt(self.idt_A, self.real_B) * lambda_B * lambda_idt
# G_B should be identity if real_A is fed: ||G_B(A) - A||
self.idt_B = self.netG_B(self.real_A)
self.loss_idt_B = self.criterionIdt(self.idt_B, self.real_A) * lambda_A * lambda_idt
else:
self.loss_idt_A = 0
self.loss_idt_B = 0
# GAN loss D_A(G_A(A))
self.loss_G_A = self.criterionGAN(self.netD_A(self.fake_B), True)
# GAN loss D_B(G_B(B))
self.loss_G_B = self.criterionGAN(self.netD_B(self.fake_A), True)
# Forward cycle loss || G_B(G_A(A)) - A||
self.loss_cycle_A = self.criterionCycle(self.rec_A, self.real_A) * lambda_A
# Backward cycle loss || G_A(G_B(B)) - B||
self.loss_cycle_B = self.criterionCycle(self.rec_B, self.real_B) * lambda_B
# combined loss and calculate gradients
self.loss_G = self.loss_G_A + self.loss_G_B + self.loss_cycle_A + self.loss_cycle_B + self.loss_idt_A + self.loss_idt_B
self.loss_G.backward()
def optimize_parameters(self):
"""Calculate losses, gradients, and update network weights; called in every training iteration"""
# forward
self.forward() # compute fake images and reconstruction images.
# G_A and G_B
self.set_requires_grad([self.netD_A, self.netD_B], False) # Ds require no gradients when optimizing Gs
self.optimizer_G.zero_grad() # set G_A and G_B's gradients to zero
self.backward_G() # calculate gradients for G_A and G_B
self.optimizer_G.step() # update G_A and G_B's weights
# D_A and D_B
self.set_requires_grad([self.netD_A, self.netD_B], True)
self.optimizer_D.zero_grad() # set D_A and D_B's gradients to zero
self.backward_D_A() # calculate gradients for D_A
self.backward_D_B() # calculate graidents for D_B
self.optimizer_D.step() # update D_A and D_B's weights

1219
RefineDNet/models/networks.py Normal file
View File

File diff suppressed because it is too large Load Diff

127
RefineDNet/models/pix2pix_model.py Normal file
View File

@@ -0,0 +1,127 @@
import torch
from .base_model import BaseModel
from . import networks
class Pix2PixModel(BaseModel):
""" This class implements the pix2pix model, for learning a mapping from input images to output images given paired data.
The model training requires '--dataset_mode aligned' dataset.
By default, it uses a '--netG unet256' U-Net generator,
a '--netD basic' discriminator (PatchGAN),
and a '--gan_mode' vanilla GAN loss (the cross-entropy objective used in the orignal GAN paper).
pix2pix paper: https://arxiv.org/pdf/1611.07004.pdf
"""
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new dataset-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
For pix2pix, we do not use image buffer
The training objective is: GAN Loss + lambda_L1 * ||G(A)-B||_1
By default, we use vanilla GAN loss, UNet with batchnorm, and aligned datasets.
"""
# changing the default values to match the pix2pix paper (https://phillipi.github.io/pix2pix/)
parser.set_defaults(norm='batch', netG='unet_256', dataset_mode='aligned')
if is_train:
parser.set_defaults(pool_size=0, gan_mode='vanilla')
parser.add_argument('--lambda_L1', type=float, default=100.0, help='weight for L1 loss')
return parser
def __init__(self, opt):
"""Initialize the pix2pix class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
BaseModel.__init__(self, opt)
# specify the training losses you want to print out. The training/test scripts will call <BaseModel.get_current_losses>
self.loss_names = ['G_GAN', 'G_L1', 'D_real', 'D_fake']
# specify the images you want to save/display. The training/test scripts will call <BaseModel.get_current_visuals>
self.visual_names = ['real_A', 'fake_B', 'real_B']
# specify the models you want to save to the disk. The training/test scripts will call <BaseModel.save_networks> and <BaseModel.load_networks>
if self.isTrain:
self.model_names = ['G', 'D']
else: # during test time, only load G
self.model_names = ['G']
# define networks (both generator and discriminator)
self.netG = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netG, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain: # define a discriminator; conditional GANs need to take both input and output images; Therefore, #channels for D is input_nc + output_nc
self.netD = networks.define_D(opt.input_nc + opt.output_nc, opt.ndf, opt.netD,
opt.n_layers_D, opt.norm, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain:
# define loss functions
self.criterionGAN = networks.GANLoss(opt.gan_mode).to(self.device)
self.criterionL1 = torch.nn.L1Loss()
# initialize optimizers; schedulers will be automatically created by function <BaseModel.setup>.
self.optimizer_G = torch.optim.Adam(self.netG.parameters(), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizer_D = torch.optim.Adam(self.netD.parameters(), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizers.append(self.optimizer_G)
self.optimizers.append(self.optimizer_D)
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input (dict): include the data itself and its metadata information.
The option 'direction' can be used to swap images in domain A and domain B.
"""
AtoB = self.opt.direction == 'AtoB'
self.real_A = input['A' if AtoB else 'B'].to(self.device)
self.real_B = input['B' if AtoB else 'A'].to(self.device)
self.image_paths = input['A_paths' if AtoB else 'B_paths']
def forward(self):
"""Run forward pass; called by both functions <optimize_parameters> and <test>."""
self.fake_B = self.netG(self.real_A) # G(A)
def backward_D(self):
"""Calculate GAN loss for the discriminator"""
# Fake; stop backprop to the generator by detaching fake_B
fake_AB = torch.cat((self.real_A, self.fake_B), 1) # we use conditional GANs; we need to feed both input and output to the discriminator
pred_fake = self.netD(fake_AB.detach())
self.loss_D_fake = self.criterionGAN(pred_fake, False)
# Real
real_AB = torch.cat((self.real_A, self.real_B), 1)
pred_real = self.netD(real_AB)
self.loss_D_real = self.criterionGAN(pred_real, True)
# combine loss and calculate gradients
self.loss_D = (self.loss_D_fake + self.loss_D_real) * 0.5
self.loss_D.backward()
def backward_G(self):
"""Calculate GAN and L1 loss for the generator"""
# First, G(A) should fake the discriminator
fake_AB = torch.cat((self.real_A, self.fake_B), 1)
pred_fake = self.netD(fake_AB)
self.loss_G_GAN = self.criterionGAN(pred_fake, True)
# Second, G(A) = B
self.loss_G_L1 = self.criterionL1(self.fake_B, self.real_B) * self.opt.lambda_L1
# combine loss and calculate gradients
self.loss_G = self.loss_G_GAN + self.loss_G_L1
self.loss_G.backward()
def optimize_parameters(self):
self.forward() # compute fake images: G(A)
# update D
self.set_requires_grad(self.netD, True) # enable backprop for D
self.optimizer_D.zero_grad() # set D's gradients to zero
self.backward_D() # calculate gradients for D
self.optimizer_D.step() # update D's weights
# update G
self.set_requires_grad(self.netD, False) # D requires no gradients when optimizing G
self.optimizer_G.zero_grad() # set G's gradients to zero
self.backward_G() # calculate graidents for G
self.optimizer_G.step() # udpate G's weights

226
RefineDNet/models/refined_DCP_model.py Normal file
View File

@@ -0,0 +1,226 @@
import torch
import itertools
from util.image_pool import ImagePool
from .base_model import BaseModel
from . import networks
import torch.nn.functional as F
from util import util
class RefinedDCPModel(BaseModel):
"""
This class implements the RefineDNet model, for learning single image dehazing without paired data.
It adopts the basic backbone networks provided by CycleGAN.
The model training requires '--dataset_mode unpaired' dataset.
By default, it uses a '--netR_T unet_trans_256' U-Net refiner,
a '--netR_J resnet_9blocks' ResNet refiner,
and a '--netD basic' discriminator (PatchGAN introduced by pix2pix).
"""
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new dataset-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
"""
parser.set_defaults(no_dropout=True) # default CycleGAN did not use dropout
if is_train:
parser.add_argument('--lambda_G', type=float, default=0.05, help='weight for loss_G_single')
parser.add_argument('--lambda_identity', type=float, default=1, help='use identity mapping. Setting lambda_identity other than 0 has an effect of scaling the weight of the identity mapping loss. For example, if the weight of the identity loss should be 10 times smaller than the weight of the reconstruction loss, please set lambda_identity = 0.1')
parser.add_argument('--lambda_rec_I', type=float, default=1, help='weight for loss_rec_I')
parser.add_argument('--lambda_tv', type=float, default=1, help='weight for TV loss of refine_T')
parser.add_argument('--lambda_vgg', type=float, default=0, help='weight for loss_vgg')
parser.add_argument('--netR_T', type=str, default='unet_trans_256', help='specify generator architecture')
parser.add_argument('--netR_J', type=str, default='resnet_9blocks', help='specify generator architecture')
return parser
def __init__(self, opt):
"""Initialize the RefineDNet class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
BaseModel.__init__(self, opt)
# specify the training losses you want to print out. The training/test scripts will call <BaseModel.get_current_losses>
self.loss_names = ['D_single', 'G_single', 'rec_I', 'TV_T', 'idt_J', 'vgg']
# specify the images you want to save/display. The training/test scripts will call <BaseModel.get_current_visuals>
if self.isTrain:
self.visual_names = ['real_I', 'dcp_T_vis', 'refine_T_vis', 'out_T_vis', 'dcp_J','refine_J', 'rec_I', 'rec_J','map_A', 'real_J', 'ref_real_J']
else:
self.visual_names = ['real_I', 'dcp_T_vis', 'refine_T_vis', 'out_T_vis', 'dcp_J','refine_J', 'rec_I', 'rec_J','map_A']
# specify the models you want to save to the disk. The training/test scripts will call <BaseModel.save_networks> and <BaseModel.load_networks>.
if self.isTrain:
self.model_names = ['Refiner_T', 'Refiner_J', 'D']
else: # during test time, only load Gs
self.model_names = ['Refiner_T', 'Refiner_J']
# define networks (both Generators and discriminators)
self.netG_DCP = networks.init_net(networks.DCPDehazeGenerator(), gpu_ids=self.gpu_ids) # use default setting for DCP
self.netRefiner_T = networks.define_G(opt.input_nc+1, 1, opt.ngf, opt.netR_T, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
self.netRefiner_J = networks.define_G(opt.input_nc+opt.output_nc, opt.output_nc, opt.ngf, opt.netR_J, opt.norm,
not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain: # define discriminators
self.netD = networks.define_D(opt.input_nc, opt.ndf, opt.netD,
opt.n_layers_D, opt.norm, opt.init_type, opt.init_gain, self.gpu_ids)
if self.isTrain:
if opt.lambda_identity > 0.0: # only works when input and output images have the same number of channels
assert(opt.input_nc == opt.output_nc)
self.fake_I_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images
self.fake_J_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images
# # define loss functions
self.criterionGAN = networks.GANLoss(opt.gan_mode).to(self.device) # define GAN loss.
self.criterionRec = torch.nn.L1Loss()
self.criterionIdt = torch.nn.L1Loss()
self.criterionTV = networks.TVLoss()
self.criterionVGG = networks.VGGLoss() if self.opt.lambda_vgg > 0.0 else None
# initialize optimizers; schedulers will be automatically created by function <BaseModel.setup>.
self.optimizer_G = torch.optim.Adam(itertools.chain(self.netRefiner_T.parameters(), self.netRefiner_J.parameters()), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizer_D = torch.optim.Adam(self.netD.parameters(), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizers.append(self.optimizer_G)
self.optimizers.append(self.optimizer_D)
# display the architecture of each part
# print(self.netRefiner_T)
# print(self.netRefiner_J)
# if self.isTrain:
# print(self.netD)
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input (dict): include the data itself and its metadata information.
"""
self.real_I = input['haze'].to(self.device) # [-1, 1]
self.image_paths = input['paths']
if self.isTrain:
self.real_J = input['clear'].to(self.device) # [-1, 1]
def forward(self):
"""Run forward pass; called by both functions <optimize_parameters> and <test>."""
dcp_J, self.dcp_T, self.dcp_A = self.netG_DCP(self.real_I)
#scale to [-1,1]
self.dcp_J = (torch.clamp(dcp_J,0,1)-0.5)/0.5
# output scale [0,1]
self.refine_T, self.out_T = self.netRefiner_T(torch.cat((self.real_I, self.dcp_T), 1))
self.refine_J = self.netRefiner_J(torch.cat((self.real_I, self.dcp_J), 1))
# reconstruct haze image
shape = self.refine_J.shape
dcp_A_scale = self.dcp_A
self.map_A = (dcp_A_scale).reshape((1,3,1,1)).repeat(1,1,shape[2],shape[3])
refine_T_map = self.refine_T.repeat(1,3,1,1)
self.rec_I = util.synthesize_fog(self.refine_J, refine_T_map, self.map_A)
self.rec_J = util.reverse_fog(self.real_I, refine_T_map, self.map_A)
def test(self):
"""Forward function used in test time.
This function wraps <forward> function in no_grad() so we don't save intermediate steps for backprop
It also calls <compute_visuals> to produce additional visualization results
"""
with torch.no_grad():
self.forward()
self.compute_visuals()
def compute_visuals(self):
"""Calculate additional output images for visdom and HTML visualization"""
# rescale to [-1,1] for visdom
self.refine_T_vis = (self.refine_T - 0.5)/0.5
self.out_T_vis = (self.out_T - 0.5)/0.5
self.dcp_T_vis = (self.dcp_T - 0.5)/0.5
# self.map_A_vis = (self.map_A - 0.5)/0.5
def backward_D_basic(self, netD, real, fake):
"""Calculate GAN loss for the discriminator
Parameters:
netD (network) -- the discriminator D
real (tensor array) -- real images
fake (tensor array) -- images generated by a generator
Return the discriminator loss.
We also call loss_D.backward() to calculate the gradients.
"""
# Real
pred_real = netD(real)
loss_D_real = self.criterionGAN(pred_real, True)
# Fake
pred_fake = netD(fake.detach())
loss_D_fake = self.criterionGAN(pred_fake, False)
# Combined loss and calculate gradients
loss_D = (loss_D_real + loss_D_fake) * 0.5
loss_D.backward()
return loss_D
def backward_D(self):
fake_J = self.fake_I_pool.query(self.refine_J)
self.loss_D_single = self.backward_D_basic(self.netD, self.real_J, fake_J)
def backward_G(self):
lambda_idt = self.opt.lambda_identity
lambda_tv = self.opt.lambda_tv
lambda_G = self.opt.lambda_G
lambda_rec_I = self.opt.lambda_rec_I
lambda_vgg = self.opt.lambda_vgg
# Generator losses for rec_I and refine_J
self.loss_G_single = self.criterionGAN(self.netD(self.refine_J), True)*lambda_G
# Reconstrcut loss
self.loss_rec_I = self.criterionRec(self.rec_I, self.real_I) * lambda_rec_I
# perecptual loss
self.loss_vgg = self.criterionVGG(self.refine_J, self.dcp_J)*lambda_vgg if lambda_vgg > 0.0 else 0
# TV loss
self.loss_TV_T = self.criterionTV(self.out_T)*lambda_tv if lambda_tv > 0.0 else 0
# Identity loss, ||refiner_J(real_J) - real_J||
self.ref_real_J = self.netRefiner_J(torch.cat((self.real_I, self.real_J), 1))
self.loss_idt_J = self.criterionIdt(self.ref_real_J, self.real_J)*lambda_idt \
if lambda_idt > 0.0 \
else 0
self.loss_G = self.loss_G_single + self.loss_rec_I + self.loss_idt_J \
+ self.loss_TV_T \
+ self.loss_vgg
self.loss_G.backward()
def optimize_parameters(self):
"""Calculate losses, gradients, and update network weights; called in every training iteration"""
# forward
self.forward() # compute fake images and reconstruction images.
# G_A and G_B
self.set_requires_grad(self.netD, False) # Ds require no gradients when optimizing Gs
self.optimizer_G.zero_grad() # set G_A and G_B's gradients to zero
self.backward_G() # calculate gradients for G_A and G_B
self.optimizer_G.step() # update G_A and G_B's weights
# D_A and D_B
self.set_requires_grad(self.netD, True)
self.optimizer_D.zero_grad() # set D_A and D_B's gradients to zero
self.backward_D() # calculate gradients for D_A
self.optimizer_D.step() # update D_A and D_B's weights

99
RefineDNet/models/template_model.py Normal file
View File

@@ -0,0 +1,99 @@
"""Model class template
This module provides a template for users to implement custom models.
You can specify '--model template' to use this model.
The class name should be consistent with both the filename and its model option.
The filename should be <model>_dataset.py
The class name should be <Model>Dataset.py
It implements a simple image-to-image translation baseline based on regression loss.
Given input-output pairs (data_A, data_B), it learns a network netG that can minimize the following L1 loss:
min_<netG> ||netG(data_A) - data_B||_1
You need to implement the following functions:
<modify_commandline_options>:　Add model-specific options and rewrite default values for existing options.
<__init__>: Initialize this model class.
<set_input>: Unpack input data and perform data pre-processing.
<forward>: Run forward pass. This will be called by both <optimize_parameters> and <test>.
<optimize_parameters>: Update network weights; it will be called in every training iteration.
"""
import torch
from .base_model import BaseModel
from . import networks
class TemplateModel(BaseModel):
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new model-specific options and rewrite default values for existing options.
Parameters:
parser -- the option parser
is_train -- if it is training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
"""
parser.set_defaults(dataset_mode='aligned') # You can rewrite default values for this model. For example, this model usually uses aligned dataset as its dataset.
if is_train:
parser.add_argument('--lambda_regression', type=float, default=1.0, help='weight for the regression loss') # You can define new arguments for this model.
return parser
def __init__(self, opt):
"""Initialize this model class.
Parameters:
opt -- training/test options
A few things can be done here.
- (required) call the initialization function of BaseModel
- define loss function, visualization images, model names, and optimizers
"""
BaseModel.__init__(self, opt) # call the initialization method of BaseModel
# specify the training losses you want to print out. The program will call base_model.get_current_losses to plot the losses to the console and save them to the disk.
self.loss_names = ['loss_G']
# specify the images you want to save and display. The program will call base_model.get_current_visuals to save and display these images.
self.visual_names = ['data_A', 'data_B', 'output']
# specify the models you want to save to the disk. The program will call base_model.save_networks and base_model.load_networks to save and load networks.
# you can use opt.isTrain to specify different behaviors for training and test. For example, some networks will not be used during test, and you don't need to load them.
self.model_names = ['G']
# define networks; you can use opt.isTrain to specify different behaviors for training and test.
self.netG = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netG, gpu_ids=self.gpu_ids)
if self.isTrain: # only defined during training time
# define your loss functions. You can use losses provided by torch.nn such as torch.nn.L1Loss.
# We also provide a GANLoss class "networks.GANLoss". self.criterionGAN = networks.GANLoss().to(self.device)
self.criterionLoss = torch.nn.L1Loss()
# define and initialize optimizers. You can define one optimizer for each network.
# If two networks are updated at the same time, you can use itertools.chain to group them. See cycle_gan_model.py for an example.
self.optimizer = torch.optim.Adam(self.netG.parameters(), lr=opt.lr, betas=(opt.beta1, 0.999))
self.optimizers = [self.optimizer]
# Our program will automatically call <model.setup> to define schedulers, load networks, and print networks
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input: a dictionary that contains the data itself and its metadata information.
"""
AtoB = self.opt.direction == 'AtoB' # use <direction> to swap data_A and data_B
self.data_A = input['A' if AtoB else 'B'].to(self.device) # get image data A
self.data_B = input['B' if AtoB else 'A'].to(self.device) # get image data B
self.image_paths = input['A_paths' if AtoB else 'B_paths'] # get image paths
def forward(self):
"""Run forward pass. This will be called by both functions <optimize_parameters> and <test>."""
self.output = self.netG(self.data_A) # generate output image given the input data_A
def backward(self):
"""Calculate losses, gradients, and update network weights; called in every training iteration"""
# caculate the intermediate results if necessary; here self.output has been computed during function <forward>
# calculate loss given the input and intermediate results
self.loss_G = self.criterionLoss(self.output, self.data_B) * self.opt.lambda_regression
self.loss_G.backward() # calculate gradients of network G w.r.t. loss_G
def optimize_parameters(self):
"""Update network weights; it will be called in every training iteration."""
self.forward() # first call forward to calculate intermediate results
self.optimizer.zero_grad() # clear network G's existing gradients
self.backward() # calculate gradients for network G
self.optimizer.step() # update gradients for network G

69
RefineDNet/models/test_model.py Normal file
View File

@@ -0,0 +1,69 @@
from .base_model import BaseModel
from . import networks
class TestModel(BaseModel):
""" This TesteModel can be used to generate CycleGAN results for only one direction.
This model will automatically set '--dataset_mode single', which only loads the images from one collection.
See the test instruction for more details.
"""
@staticmethod
def modify_commandline_options(parser, is_train=True):
"""Add new dataset-specific options, and rewrite default values for existing options.
Parameters:
parser -- original option parser
is_train (bool) -- whether training phase or test phase. You can use this flag to add training-specific or test-specific options.
Returns:
the modified parser.
The model can only be used during test time. It requires '--dataset_mode single'.
You need to specify the network using the option '--model_suffix'.
"""
assert not is_train, 'TestModel cannot be used during training time'
parser.set_defaults(dataset_mode='single')
parser.add_argument('--model_suffix', type=str, default='', help='In checkpoints_dir, [epoch]_net_G[model_suffix].pth will be loaded as the generator.')
return parser
def __init__(self, opt):
"""Initialize the pix2pix class.
Parameters:
opt (Option class)-- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
assert(not opt.isTrain)
BaseModel.__init__(self, opt)
# specify the training losses you want to print out. The training/test scripts will call <BaseModel.get_current_losses>
self.loss_names = []
# specify the images you want to save/display. The training/test scripts will call <BaseModel.get_current_visuals>
self.visual_names = ['real', 'fake']
# specify the models you want to save to the disk. The training/test scripts will call <BaseModel.save_networks> and <BaseModel.load_networks>
self.model_names = ['G' + opt.model_suffix] # only generator is needed.
self.netG = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netG,
opt.norm, not opt.no_dropout, opt.init_type, opt.init_gain, self.gpu_ids)
# assigns the model to self.netG_[suffix] so that it can be loaded
# please see <BaseModel.load_networks>
setattr(self, 'netG' + opt.model_suffix, self.netG) # store netG in self.
def set_input(self, input):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input: a dictionary that contains the data itself and its metadata information.
We need to use 'single_dataset' dataset mode. It only load images from one domain.
"""
self.real = input['A'].to(self.device)
self.image_paths = input['A_paths']
def forward(self):
"""Run forward pass."""
self.fake = self.netG(self.real) # G(real)
def optimize_parameters(self):
"""No optimization for test model."""
pass