def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
self.ssim_loss = settings.ssim_loss
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
ensure_dir('../log_test')
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
if len(settings.device_id) > 1:
self.net = nn.DataParallel(ODE_DerainNet()).cuda()
else:
torch.cuda.set_device(settings.device_id[0])
self.net = ODE_DerainNet().cuda()
self.l1 = nn.L1Loss().cuda()
self.mse = nn.MSELoss().cuda()
self.ssim = SSIM().cuda()
self.step = 0
self.save_steps = settings.save_steps
self.num_workers = settings.num_workers
self.batch_size = settings.batch_size
self.writers = {}
self.dataloaders = {}
self.opt_net = Adam(self.net.parameters(), lr=settings.lr)
self.sche_net = MultiStepLR(self.opt_net,
milestones=[settings.l1, settings.l2],
gamma=0.1)
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
self.net = TFN().cuda()
self.crit = L1Loss().cuda()
self.ssim = SSIM().cuda()
self.msssim = MSSSIM().cuda()
self.step = 0
self.perceptual_weight = settings.perceptual_weight
self.loss_weight = settings.loss_weight
self.total_variation_weight = settings.total_variation_weight
self.ssim_loss_weight = settings.ssim_loss_weight
self.save_steps = settings.save_steps
self.num_workers = settings.num_workers
self.batch_size = settings.batch_size
self.writers = {}
self.dataloaders = {}
self.opt = Adam(self.net.parameters(), lr=settings.lr)
self.sche = MultiStepLR(
self.opt,
milestones=[11000, 70000, 90000, 110000, 130000],
gamma=0.1)
def __init__(self):
self.device = torch.device("cuda")
self.log_dir = './logdir'
self.model_dir = './model'
ensure_dir(self.log_dir)
ensure_dir(self.model_dir)
self.log_name = 'train_derain'
self.val_log_name = 'val_derain'
logger.info('set log dir as %s' % self.log_dir)
logger.info('set model dir as %s' % self.model_dir)
self.test_data_path = 'testing/real_test_1000.txt' # test dataset txt file path
self.train_data_path = 'training/real_world.txt' # train dataset txt file path
self.multi_gpu = False
torch.cuda.empty_cache()
self.net = SPANet().to(self.device)
print_network(self.net)
self.l1 = nn.L1Loss().to(self.device)
self.l2 = nn.MSELoss().to(self.device)
self.ssim = SSIM().to(self.device)
self.step = 0
self.save_steps = 400
self.num_workers = 1
self.batch_size = 2
self.writers = {}
self.dataloaders = {}
self.shuffle = True
self.opt = Adam(self.net.parameters(), lr=5e-3)
self.sche = MultiStepLR(self.opt, milestones=[500, 1500,2000,3000], gamma=0.1)
self.step_time = 0
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
# self.net = RESCAN().cuda()
if len(settings.device_id) >1:
self.net = nn.DataParallel(RESCAN()).cuda()
else:
self.net = RESCAN().cuda()
self.crit = MSELoss().cuda()
self.ssim = SSIM().cuda()
self.step = 0
self.save_steps = settings.save_steps
self.num_workers = settings.num_workers
self.batch_size = settings.batch_size
self.writers = {}
self.dataloaders = {}
self.opt = Adam(self.net.parameters(), lr=settings.lr)
self.sche = MultiStepLR(self.opt, milestones=[240000, 320000], gamma=0.1)
def __init__(self):
self.show_dir = settings.show_dir
self.model_dir = settings.model_dir
ensure_dir(settings.show_dir)
ensure_dir(settings.model_dir)
logger.info('set show dir as %s' % settings.show_dir)
logger.info('set model dir as %s' % settings.model_dir)
if len(settings.device_id) > 1:
self.net = nn.DataParallel(Net()).cuda()
else:
self.net = Net().cuda()
self.ssim = SSIM().cuda()
self.dataloaders = {}
self.ssim = SSIM().cuda()
self.a = 0
self.t = 0
def __init__(self):
self.show_dir = settings.show_dir
self.model_dir = settings.model_dir
ensure_dir(settings.show_dir)
ensure_dir(settings.model_dir)
logger.info('set show dir as %s' % settings.show_dir)
logger.info('set model dir as %s' % settings.model_dir)
self.net = RESCAN().cuda()
self.dataloaders = {}
self.ssim=SSIM().cuda()
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
self.net = RESCAN().cuda()
self.crit = MSELoss().cuda()
self.ssim = SSIM().cuda()
self.dataloaders = {}
def __init__(self):
self.show_dir = settings.show_dir
self.model_dir = settings.model_dir
ensure_dir(settings.show_dir)
ensure_dir(settings.model_dir)
logger.info('set show dir as %s' % settings.show_dir)
logger.info('set model dir as %s' % settings.model_dir)
if len(settings.device_id) > 1:
self.net = nn.DataParallel(ODE_DerainNet()).cuda()
#self.l2 = nn.DataParallel(MSELoss(),settings.device_id)
#self.l1 = nn.DataParallel(nn.L1Loss(),settings.device_id)
#self.ssim = nn.DataParallel(SSIM(),settings.device_id)
#self.vgg = nn.DataParallel(VGG(),settings.device_id)
else:
torch.cuda.set_device(settings.device_id[0])
self.net = ODE_DerainNet().cuda()
self.ssim = SSIM().cuda()
self.dataloaders = {}
self.ssim = SSIM().cuda()
self.a = 0
self.t = 0
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
if len(settings.device_id) >1:
self.net = nn.DataParallel(Net()).cuda()
else:
self.net = Net().cuda()
self.l2 = MSELoss().cuda()
self.l1 = nn.L1Loss().cuda()
self.ssim = SSIM().cuda()
self.dataloaders = {}
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
ensure_dir('../log_test')
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
if torch.cuda.is_available():
self.net = Net().cuda()
self.dis_rain_img = Discriminator_rain_img().cuda()
self.dis_img = Discriminator_img().cuda()
if len(device_ids) > 1:
self.net = nn.DataParallel(Net()).cuda()
self.dis_rain_img = nn.DataParallel(
Discriminator_rain_img()).cuda()
self.dis_img = nn.DataParallel(Discriminator_img()).cuda()
self.l2 = MSELoss().cuda()
self.l1 = nn.L1Loss().cuda()
self.ssim = SSIM().cuda()
# self.vgg = VGG().cuda()
self.bceloss = nn.BCELoss().cuda()
self.step = 0
self.ssim_val = 0
self.psnr_val = 0
self.save_steps = settings.save_steps
self.num_workers = settings.num_workers
self.batch_size = settings.batch_size
self.writers = {}
self.dataloaders = {}
self.opt_net = Adam(self.net.parameters(), lr=settings.lr)
self.sche_net = MultiStepLR(self.opt_net,
milestones=[settings.l1, settings.l2],
gamma=0.1)
self.opt_dis_rain_img = Adam(self.dis_rain_img.parameters(),
lr=settings.lr)
self.sche_dis_rain_img = MultiStepLR(
self.opt_dis_rain_img,
milestones=[settings.l1, settings.l2],
gamma=0.1)
self.opt_dis_img = Adam(self.dis_img.parameters(), lr=settings.lr)
self.sche_dis_img = MultiStepLR(self.opt_dis_img,
milestones=[settings.l1, settings.l2],
gamma=0.1)
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
self.net = DetailNet().cuda()
self.crit = MSELoss().cuda()
self.ssim = SSIM().cuda()
self.step = 0
self.save_steps = settings.save_steps
self.num_workers = settings.num_workers
self.batch_size = settings.batch_size
self.writers = {}
self.dataloaders = {}
self.opt = Adam(self.net.parameters(), lr=settings.lr)
self.sche = MultiStepLR(self.opt, milestones=[15000, 17500], gamma=0.1)
def __init__(self, args):
self.log_dir = args.log_dir
self.model_dir = args.model_dir
ensure_dir(self.log_dir)
ensure_dir(self.model_dir)
logger.info('set log dir as %s' % self.log_dir)
logger.info('set model dir as %s' % self.model_dir)
self.net = ERLNet(in_channels=3, out_channels=3).cuda()
self.ssim = SSIM().cuda()
self.image_size = args.image_size
self.batch_size = args.batch_size
self.num_workers = args.num_workers
self.step = 0
self.epoch = args.epochs
self.now_epoch = 0
self.start_epoch = 0
self.writers = {}
self.total_step = 0
self.sessname = args.sessname
def __init__(self, args):
self.log_dir = args.log_dir
self.model_dir = args.model_dir
ensure_dir(self.log_dir)
ensure_dir(self.model_dir)
logger.info('set log dir as %s' % self.log_dir)
logger.info('set model dir as %s' % self.model_dir)
self.net = detail_net().cuda()
self.ssim = SSIM().cuda()
self.image_size = args.image_size
self.batch_size = args.batch_size
self.num_workers = args.num_workers
self.step = 0
self.epoch = args.epochs
self.now_epoch = 0
self.start_epoch = 0
self.writers = {}
self.total_step = 0
self.sessname = args.sessname
if args.loss == "MSE":
self.crit = MSELoss().cuda()
elif args.loss == "L1Loss":
self.crit = L1Loss().cuda()
else:
self.crit = Myloss().cuda()
if args.opt == "SGD":
self.opt = SGD(self.net.parameters(), lr=args.lr)
else:
self.opt = Adam(self.net.parameters(), lr=args.lr)
self.sche = MultiStepLR(self.opt,
milestones=[30, 60, 90, 120, 150, 180, 210],
gamma=0.5)
def __init__(self):
self.log_dir = settings.log_dir
self.model_dir = settings.model_dir
ensure_dir(settings.log_dir)
ensure_dir(settings.model_dir)
logger.info('set log dir as %s' % settings.log_dir)
logger.info('set model dir as %s' % settings.model_dir)
if torch.cuda.is_available():
self.net = Net().cuda()
self.dis_rain_img = Discriminator_rain_img().cuda()
self.dis_img = Discriminator_img().cuda()
if len(device_ids) > 1:
self.net = nn.DataParallel(Net()).cuda()
self.dis_rain_img = nn.DataParallel(
Discriminator_rain_img()).cuda()
self.dis_img = nn.DataParallel(Discriminator_img()).cuda()
self.opt_net = Adam(self.net.parameters(), lr=settings.lr)
self.sche_net = MultiStepLR(self.opt_net,
milestones=[settings.l1, settings.l2],
gamma=0.1)
self.opt_dis_rain_img = Adam(self.dis_rain_img.parameters(),
lr=settings.lr)
self.sche_dis_rain_img = MultiStepLR(
self.opt_dis_rain_img,
milestones=[settings.l1, settings.l2],
gamma=0.1)
self.opt_dis_img = Adam(self.dis_rain_img.parameters(), lr=settings.lr)
self.sche_dis_img = MultiStepLR(self.opt_dis_rain_img,
milestones=[settings.l1, settings.l2],
gamma=0.1)
self.l2 = MSELoss().cuda()
self.l1 = nn.L1Loss().cuda()
self.ssim = SSIM().cuda()
self.vgg = VGG().cuda()
self.dataloaders = {}
def main():
print('Loading dataset ...\n')
dataset_train = Dataset(data_path='datasets_arlo/')
loader_train = DataLoader(dataset=dataset_train)
print('# of training samples :', int(len(loader_train)))
# define some hyper-parameters
recurr_iter = 4
use_GPU = True
model_path = 'logs/real/latest.pth'
num_epochs = 2
#model = PRN(recurr_iter, use_GPU)
#model= Generator_lstm(recurr_iter, use_GPU)
torch.cuda.empty_cache()
device = torch.device("cuda")
model = SPANet().to(device)
print_network(model)
model.load_state_dict(torch.load(model_path))
#loss
L1 = nn.L1Loss()
L2 = nn.MSELoss()
#binary_cross_entropy = F.binary_cross_entropy
criterion = SSIM()
if use_GPU:
model = model.cuda()
L1.cuda()
L2.cuda()
criterion.cuda()
#binary_cross_entropy.cuda()
#optimizer:
optimizer = optim.Adam(model.parameters(), lr=1e-4)
scheduler = MultiStepLR(optimizer, milestones=[20, 40], gamma=0.1)
#record training
writer = SummaryWriter('logs/')
step = 0
for epoch in range(num_epochs):
print('Epoch {}/{}'.format(epoch, num_epochs - 1))
print('-' * 10)
for i, (input_train, target_train,
streak_train) in enumerate(loader_train, 0):
start = timeit.default_timer()
input_train, target_train, streak_train = Variable(
input_train,
requires_grad=False), Variable(target_train,
requires_grad=False), Variable(
streak_train,
requires_grad=False)
if use_GPU:
input_train, target_train, streak_train = input_train.cuda(
), target_train.cuda(), streak_train.cuda()
optimizer.zero_grad()
model.train()
mask, out_train = model(input_train)
#out_train = input_train - out_streak
#out_streak= torch.clamp(out_streak[:,:,:,:], 0., 1.)
l1 = L1(mask[:, 0, :, :], streak_train[:, 0, :, :])
l2 = L2(streak_train[:, 0, :, :], mask[:, 0, :, :])
ssim = criterion(target_train, out_train)
pixel_metric = l1 + l2 + (
1 - ssim
) #L2(streak_train[:,0,:,:],mask[:,0,:,:]) + L1(mask[:,0,:,:], streak_train[:,0,:,:])+ (1- criterion(target_train,out_train)) #L1(streak_train[:,0,:,:], out_streak[:,0,:,:]) +
#loss= -pixel_metric
loss = pixel_metric
loss.backward()
optimizer.step()
model.eval()
mask, out_train = model(input_train)
stop = timeit.default_timer()
print(
"[epoch %d][%d/%d] loss: %.4f, l1 loss: %.4f, l2 loss: %.4f, ssim: %.4f, step time: %.2f"
% (epoch + 1, i + 1, len(loader_train), loss.item(), l1.item(),
l2.item(), ssim.item(), stop - start))
if step % 10 == 0:
writer.add_scalar('loss', loss.item(), step)
step += 1
model.eval()
mask, out_train = model(input_train)
im_target = utils.make_grid(target_train.data,
nrow=8,
normalize=True,
scale_each=True)
im_input = utils.make_grid(input_train.data,
nrow=8,
normalize=True,
scale_each=True)
out_target = utils.make_grid(out_train.data,
nrow=8,
normalize=True,
scale_each=True)
writer.add_image('clean image', im_target, epoch + 1)
writer.add_image('rainy image', im_input, epoch + 1)
writer.add_image('streak image', out_target, epoch + 1)
torch.save(model.state_dict(), 'logs/real/latest.pth')