changes

face_swapper/src/models/generator.py

@@ -1,10 +1,8 @@
 from configparser import ConfigParser
-from typing import Tuple
 
 from torch import Tensor, nn
 
 from ..networks.aad import AAD
-from ..networks.masknet import MaskNet
 from ..networks.unet import UNet
 from ..types import Attributes, Embedding
 
@@ -14,16 +12,13 @@ class Generator(nn.Module):
 		super().__init__()
 		self.encoder = UNet(config_parser)
 		self.generator = AAD(config_parser)
-		self.masker = MaskNet(config_parser)
 		self.encoder.apply(init_weight)
 		self.generator.apply(init_weight)
-		self.masker.apply(init_weight)
 
-	def forward(self, source_embedding : Embedding, target_tensor : Tensor) -> Tuple[Tensor, Tensor]:
+	def forward(self, source_embedding : Embedding, target_tensor : Tensor) -> Tensor:
 		target_attributes = self.get_attributes(target_tensor)
 		output_tensor = self.generator(source_embedding, target_attributes)
-		mask_tensor = self.masker(target_tensor, target_attributes[-1])
-		return output_tensor, mask_tensor
+		return output_tensor
 
 	def get_attributes(self, input_tensor : Tensor) -> Attributes:
 		return self.encoder(input_tensor)

face_swapper/src/models/loss.py

@@ -180,18 +180,16 @@ class GazeLoss(nn.Module):
 class MaskLoss(nn.Module):
 	def __init__(self, config_parser : ConfigParser, parser : ParserModule) -> None:
 		super().__init__()
-		self.config_mask_weight = config_parser.getfloat('training.losses', 'mask_weight')
 		self.config_output_size = config_parser.getint('training.model.generator', 'output_size')
 		self.parser = parser
 		self.mse_loss = nn.MSELoss()
 
-	def forward(self, target_tensor : Tensor, mask_tensor : Tensor) -> Tuple[Tensor, Tensor]:
+	def forward(self, target_tensor : Tensor, mask_tensor : Tensor) -> Tensor:
 		target_mask = self.calc_mask(target_tensor)
 		target_mask = target_mask.view(-1, self.config_output_size, self.config_output_size)
 		mask_tensor = mask_tensor.view(-1, self.config_output_size, self.config_output_size)
 		mask_loss = self.mse_loss(target_mask, mask_tensor)
-		weighted_mask_loss = mask_loss * self.config_mask_weight
-		return mask_loss, weighted_mask_loss
+		return mask_loss
 
 	def calc_mask(self, target_tensor : Tensor) -> Tensor:
 		target_tensor = torch.nn.functional.interpolate(target_tensor, (512, 512), mode = 'bilinear')

face_swapper/src/training.py

@@ -17,6 +17,7 @@ from .helper import calc_embedding
 from .models.discriminator import Discriminator
 from .models.generator import Generator
 from .models.loss import AdversarialLoss, AttributeLoss, DiscriminatorLoss, GazeLoss, IdentityLoss, MaskLoss, MotionLoss, ReconstructionLoss
+from .networks.masknet import MaskNet
 from .types import Batch, Embedding, OptimizerSet
 
 warnings.filterwarnings('ignore', category = UserWarning, module = 'torch')
@@ -40,6 +41,7 @@ class FaceSwapperTrainer(LightningModule):
 		self.parser = torch.jit.load(self.config_parser_path, map_location = 'cpu').eval()
 		self.generator = Generator(config_parser)
 		self.discriminator = Discriminator(config_parser)
+		self.masker = MaskNet(config_parser)
 		self.discriminator_loss = DiscriminatorLoss()
 		self.adversarial_loss = AdversarialLoss(config_parser)
 		self.attribute_loss = AttributeLoss(config_parser)
@@ -54,11 +56,13 @@ class FaceSwapperTrainer(LightningModule):
 		output_tensor, mask_tensor = self.generator(source_embedding, target_tensor)
 		return output_tensor, mask_tensor
 
-	def configure_optimizers(self) -> Tuple[OptimizerSet, OptimizerSet]:
+	def configure_optimizers(self) -> Tuple[OptimizerSet, OptimizerSet, OptimizerSet]:
 		generator_optimizer = torch.optim.AdamW(self.generator.parameters(), lr = self.config_learning_rate, betas = (0.0, 0.999), weight_decay = 1e-4)
 		discriminator_optimizer = torch.optim.AdamW(self.discriminator.parameters(), lr = self.config_learning_rate, betas = (0.0, 0.999), weight_decay = 1e-4)
+		masker_optimizer = torch.optim.AdamW(self.masker.parameters(), lr = self.config_learning_rate, betas = (0.0, 0.999), weight_decay = 1e-4)
 		generator_scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(generator_optimizer, T_0 = 300, T_mult = 2)
 		discriminator_scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(discriminator_optimizer, T_0 = 300, T_mult = 2)
+		masker_scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(masker_optimizer, T_0 = 300, T_mult = 2)
 
 		generator_config =\
 		{
@@ -78,14 +82,23 @@ class FaceSwapperTrainer(LightningModule):
 				'interval': 'step'
 			}
 		}
-		return generator_config, discriminator_config
+		masker_config =\
+		{
+			'optimizer': masker_optimizer,
+			'lr_scheduler':
+			{
+				'scheduler': masker_scheduler,
+				'interval': 'step'
+			}
+		}
+		return generator_config, discriminator_config, masker_config
 
 	def training_step(self, batch : Batch, batch_index : int) -> Tensor:
 		source_tensor, target_tensor = batch
-		generator_optimizer, discriminator_optimizer = self.optimizers() #type:ignore[attr-defined]
+		generator_optimizer, discriminator_optimizer, masker_optimizer = self.optimizers() #type:ignore[attr-defined]
 		source_embedding = calc_embedding(self.embedder, source_tensor, (0, 0, 0, 0))
 		target_attributes = self.generator.get_attributes(target_tensor)
-		generator_output_tensor, generator_mask_tensor = self.generator(source_embedding, target_tensor)
+		generator_output_tensor = self.generator(source_embedding, target_tensor)
 		generator_output_attributes = self.generator.get_attributes(generator_output_tensor)
 		discriminator_output_tensors = self.discriminator(generator_output_tensor)
 
@@ -96,14 +109,22 @@ class FaceSwapperTrainer(LightningModule):
 		identity_loss, weighted_identity_loss = self.identity_loss(generator_output_tensor, source_tensor)
 		pose_loss, weighted_pose_loss, expression_loss, weighted_expression_loss = self.motion_loss(target_tensor, generator_output_tensor)
 		gaze_loss, weighted_gaze_loss = self.gaze_loss(target_tensor, generator_output_tensor)
-		mask_loss, weighted_mask_loss = self.mask_loss(target_tensor, generator_mask_tensor)
-		generator_loss = weighted_adversarial_loss + weighted_attribute_loss + weighted_reconstruction_loss + weighted_identity_loss + weighted_pose_loss + weighted_gaze_loss + weighted_expression_loss + weighted_mask_loss
+		generator_loss = weighted_adversarial_loss + weighted_attribute_loss + weighted_reconstruction_loss + weighted_identity_loss + weighted_pose_loss + weighted_gaze_loss + weighted_expression_loss
 
 		generator_optimizer.zero_grad()
 		self.manual_backward(generator_loss)
 		generator_optimizer.step()
 		self.untoggle_optimizer(generator_optimizer)
 
+		self.toggle_optimizer(masker_optimizer)
+		mask_tensor = self.masker(target_tensor, target_attributes[-1].detach())
+		mask_loss = self.mask_loss(target_tensor, mask_tensor)
+
+		masker_optimizer.zero_grad()
+		self.manual_backward(mask_loss)
+		masker_optimizer.step()
+		self.untoggle_optimizer(masker_optimizer)
+
 		self.toggle_optimizer(discriminator_optimizer)
 		discriminator_source_tensors = self.discriminator(source_tensor)
 		discriminator_output_tensors = self.discriminator(generator_output_tensor.detach())
@@ -131,7 +152,7 @@ class FaceSwapperTrainer(LightningModule):
 	def validation_step(self, batch : Batch, batch_index : int) -> Tensor:
 		source_tensor, target_tensor = batch
 		source_embedding = calc_embedding(self.embedder, source_tensor, (0, 0, 0, 0))
-		output_tensor, mask_tensor = self.generator(source_embedding, target_tensor)
+		output_tensor = self.generator(source_embedding, target_tensor)
 		output_embedding = calc_embedding(self.embedder, output_tensor, (0, 0, 0, 0))
 		validation_score = (nn.functional.cosine_similarity(source_embedding, output_embedding).mean() + 1) * 0.5
 		self.log('validation_score', validation_score, prog_bar = True)