Config changes for discriminator advantage run

Still going from high->low, discriminator discerns on low. Next up disc works on high.

codes/models/SRGAN_model.py

@@ -5,7 +5,7 @@ import torch.nn as nn
 from torch.nn.parallel import DataParallel, DistributedDataParallel
 import models.networks as networks
 import models.lr_scheduler as lr_scheduler
-from .base_model import BaseModel
+from models.base_model import BaseModel
 from models.loss import GANLoss
 from apex import amp
 
@@ -150,8 +150,11 @@ class SRGANModel(BaseModel):
         for p in self.netD.parameters():
             p.requires_grad = False
 
-        self.optimizer_G.zero_grad()
+        if step > self.D_init_iters:
-        self.fake_H = self.netG(self.var_L)
+            self.optimizer_G.zero_grad()
+            self.fake_H = self.netG(self.var_L)
+        else:
+            self.fake_H = self.pix
 
         if step % 50 == 0:
             for i in range(self.var_L.shape[0]):
@@ -235,9 +238,9 @@ class SRGANModel(BaseModel):
                 self.log_dict['l_g_fea'] = l_g_fea.item()
             self.log_dict['l_g_gan'] = l_g_gan.item()
             self.log_dict['l_g_total'] = l_g_total.item()
-            self.log_dict['l_d_real'] = l_d_real.item()
+        self.log_dict['l_d_real'] = l_d_real.item()
-            self.log_dict['l_d_fake'] = l_d_fake.item()
+        self.log_dict['l_d_fake'] = l_d_fake.item()
-            self.log_dict['D_fake'] = torch.mean(pred_d_fake.detach())
+        self.log_dict['D_fake'] = torch.mean(pred_d_fake.detach())
 
     def test(self):
         self.netG.eval()

codes/models/archs/HighToLowResNet.py

@@ -22,9 +22,9 @@ class HighToLowResNet(nn.Module):
         # The first will be applied against the hi-res inputs and will have only 4 layers.
         # The second will be applied after half of the downscaling and will also have only 6 layers.
         # The final will be applied against the final resolution and will have all of the remaining layers.
-        self.trunk_hires = arch_util.make_layer(basic_block, 4)
+        self.trunk_hires = arch_util.make_layer(basic_block, 5)
-        self.trunk_medres = arch_util.make_layer(basic_block, 6)
+        self.trunk_medres = arch_util.make_layer(basic_block, 10)
-        self.trunk_lores = arch_util.make_layer(basic_block2, nb - 10)
+        self.trunk_lores = arch_util.make_layer(basic_block2, nb - 15)
 
         # downsampling
         if self.downscale == 4:

codes/options/test/test_ESRGAN_vrp.yml

@@ -23,4 +23,4 @@ network_G:
 
 #### path
 path:
-  pretrain_model_G: ../experiments/ESRGANx4_blacked_ramped_feat/models/35000_G.pth
+  pretrain_model_G: ../experiments/ESRGANx4_blacked_for_adrianna/models/19500_G.pth

codes/options/train/finetune_ESRGAN_blacked_for_adrianna.yml

@@ -41,9 +41,9 @@ network_D:
 
 #### path
 path:
-  pretrain_model_G: ../experiments/blacked_ft_G.pth
+  pretrain_model_G: ../experiments/ESRGANx4_blacked_for_adrianna/models/15500_G.pth
-  pretrain_model_D: ../experiments/blacked_ft_D.pth
+  pretrain_model_D: ../experiments/ESRGANx4_blacked_for_adrianna/models/15500_D.pth
-  resume_state: ~
+  resume_state: ../experiments/ESRGANx4_blacked_for_adrianna/training_state/15500.state
   strict_load: true
 
 #### training settings: learning rate scheme, loss
@@ -71,7 +71,7 @@ train:
   feature_weight_decay_steps: 2500
   feature_weight_minimum: !!float 1e-3
   gan_type: ragan  # gan | ragan
-  gan_weight: !!float 1e-2
+  gan_weight: !!float 1e-1
 
   D_update_ratio: 1
   D_init_iters: 0

codes/options/train/finetune_corruptGAN_adrianna.yml

@@ -1,84 +0,0 @@
-#### general settings
-name: ESRGAN_adrianna_corrupt_finetune
-use_tb_logger: true
-model: corruptgan
-distortion: downsample
-scale: 4
-gpu_ids: [0]
-amp_opt_level: O1
-
-#### datasets
-datasets:
-  train:
-    name: blacked
-    mode: downsample
-    dataroot_GT: ../datasets/blacked/train/hr
-    dataroot_LQ: ../datasets/adrianna/train/lr
-    mismatched_Data_OK: true
-    use_shuffle: true
-    n_workers: 4  # per GPU
-    batch_size: 16
-    target_size: 64
-    use_flip: false
-    use_rot: false
-    color: RGB
-  val:
-    name: blacked_val
-    mode: downsample
-    target_size: 64
-    dataroot_GT: ../datasets/blacked/val/hr
-    dataroot_LQ: ../datasets/blacked/val/lr
-
-#### network structures
-network_G:
-  which_model_G: HighToLowResNet
-  in_nc: 3
-  out_nc: 3
-  nf: 128
-  nb: 30
-network_D:
-  which_model_D: discriminator_vgg_128
-  in_nc: 3
-  nf: 96
-
-#### path
-path:
-  pretrain_model_G: ../experiments/blacked_lqprn_corrupt_G.pth
-  pretrain_model_D: ../experiments/blacked_lqprn_corrupt_D.pth
-  resume_state: ~
-  strict_load: true
-
-#### training settings: learning rate scheme, loss
-train:
-  lr_G: !!float 1e-5
-  weight_decay_G: 0
-  beta1_G: 0.9
-  beta2_G: 0.99
-  lr_D: !!float 1e-5
-  weight_decay_D: 0
-  beta1_D: 0.9
-  beta2_D: 0.99
-  lr_scheme: MultiStepLR
-
-  niter: 400000
-  warmup_iter: -1  # no warm up
-  lr_steps: [1000, 2000, 3000]
-  lr_gamma: 0.5
-
-  pixel_criterion: l1
-  pixel_weight: !!float 1e-2
-  feature_criterion: l1
-  feature_weight: 0
-  gan_type: gan  # gan | ragan
-  gan_weight: !!float 1e-1
-
-  D_update_ratio: 1
-  D_init_iters: 0
-
-  manual_seed: 10
-  val_freq: !!float 5e2
-
-#### logger
-logger:
-  print_freq: 50
-  save_checkpoint_freq: !!float 5e2

codes/options/train/train_GAN_blacked_corrupt.yml

@@ -1,5 +1,5 @@
 #### general settings
-name: ESRGAN_blacked_corrupt_lqprn
+name: corruptGAN_4k_lqprn_closeup
 use_tb_logger: true
 model: corruptgan
 distortion: downsample
@@ -12,9 +12,9 @@ datasets:
   train:
     name: blacked
     mode: downsample
-    dataroot_GT: ../datasets/blacked/train/hr
+    dataroot_GT: K:\\4k6k\\4k_closeup\\hr
-    dataroot_LQ: ../datasets/lqprn/train/lr
+    dataroot_LQ: E:\\4k6k\\adrianna\\for_training\\lr
-    mismatched_Data_OK: false
+    mismatched_Data_OK: true
     use_shuffle: true
     n_workers: 4  # per GPU
     batch_size: 16
@@ -34,12 +34,12 @@ network_G:
   which_model_G: HighToLowResNet
   in_nc: 3
   out_nc: 3
-  nf: 128
+  nf: 64
-  nb: 30
+  nb: 64
 network_D:
   which_model_D: discriminator_vgg_128
   in_nc: 3
-  nf: 96
+  nf: 64
 
 #### path
 path:
@@ -61,7 +61,7 @@ train:
 
   niter: 400000
   warmup_iter: -1  # no warm up
-  lr_steps: [1000, 2000, 3500, 5000, 6500]
+  lr_steps: [4000, 8000, 12000, 15000, 20000]
   lr_gamma: 0.5
 
   pixel_criterion: l1
@@ -71,8 +71,8 @@ train:
   gan_type: gan  # gan | ragan
   gan_weight: !!float 1e-1
 
-  D_update_ratio: 1
+  D_update_ratio: 2
-  D_init_iters: 0
+  D_init_iters: 500
 
   manual_seed: 10
   val_freq: !!float 5e2

codes/test.py

@@ -15,7 +15,7 @@ if __name__ == "__main__":
     #### options
     want_just_images = True
     parser = argparse.ArgumentParser()
-    parser.add_argument('-opt', type=str, help='Path to options YMAL file.', default='options/test/test_corrupt_vixen_adrianna.yml')
+    parser.add_argument('-opt', type=str, help='Path to options YMAL file.', default='options/test/test_ESRGAN_vrp.yml')
     opt = option.parse(parser.parse_args().opt, is_train=False)
     opt = option.dict_to_nonedict(opt)
 

codes/train.py

@@ -29,7 +29,7 @@ def init_dist(backend='nccl', **kwargs):
 def main():
     #### options
     parser = argparse.ArgumentParser()
-    parser.add_argument('-opt', type=str, help='Path to option YAML file.', default='options/train/finetune_corruptGAN_adrianna.yml')
+    parser.add_argument('-opt', type=str, help='Path to option YAML file.', default='options/train/train_GAN_blacked_corrupt.yml')
     parser.add_argument('--launcher', choices=['none', 'pytorch'], default='none',
                         help='job launcher')
     parser.add_argument('--local_rank', type=int, default=0)