base

configs/synthesizers/TAFG.yml

@@ -0,0 +1,145 @@
+name: TAFG
+engine: TAFG
+result_dir: ./result
+max_pairs: 1000000
+
+misc:
+  random_seed: 324
+
+checkpoint:
+  epoch_interval: 1 # one checkpoint every 1 epoch
+  n_saved: 2
+
+interval:
+  print_per_iteration: 10 # print once per 10 iteration
+  tensorboard:
+    scalar: 100
+    image: 2
+
+model:
+  generator:
+    _type: TAHG-Generator
+    _bn_to_sync_bn: False
+    style_in_channels: 3
+    content_in_channels: 1
+    num_blocks: 4
+  discriminator:
+    _type: MultiScaleDiscriminator
+    num_scale: 2
+    discriminator_cfg:
+      _type: base-PatchDiscriminator
+      in_channels: 3
+      base_channels: 64
+      use_spectral: True
+      need_intermediate_feature: True
+
+loss:
+  gan:
+    loss_type: hinge
+    real_label_val: 1.0
+    fake_label_val: 0.0
+    weight: 1.0
+  perceptual:
+    layer_weights:
+      "1": 0.03125
+      "6": 0.0625
+      "11": 0.125
+      "20": 0.25
+      "29": 1
+    criterion: 'L1'
+    style_loss: False
+    perceptual_loss: True
+    weight: 1
+  style:
+    layer_weights:
+      "1": 0.03125
+      "6": 0.0625
+      "11": 0.125
+      "20": 0.25
+      "29": 1
+    criterion: 'L2'
+    style_loss: True
+    perceptual_loss: False
+    weight: 0
+  fm:
+    level: 1
+    weight: 1
+  recon:
+    level: 1
+    weight: 1
+
+optimizers:
+  generator:
+    _type: Adam
+    lr: 0.0001
+    betas: [ 0, 0.9 ]
+    weight_decay: 0.0001
+  discriminator:
+    _type: Adam
+    lr: 4e-4
+    betas: [ 0, 0.9 ]
+    weight_decay: 0.0001
+
+data:
+  train:
+    scheduler:
+      start_proportion: 0.5
+      target_lr: 0
+    buffer_size: 50
+    dataloader:
+      batch_size: 256
+      shuffle: True
+      num_workers: 2
+      pin_memory: True
+      drop_last: True
+    dataset:
+      _type: GenerationUnpairedDatasetWithEdge
+      root_a: "/data/i2i/VoxCeleb2Anime/trainA"
+      root_b: "/data/i2i/VoxCeleb2Anime/trainB"
+      edges_path: "/data/i2i/VoxCeleb2Anime/edges"
+      edge_type: "hed"
+      size: [128, 128]
+      random_pair: True
+      pipeline:
+        - Load
+        - Resize:
+            size: [128, 128]
+        - ToTensor
+        - Normalize:
+            mean: [ 0.5, 0.5, 0.5 ]
+            std: [ 0.5, 0.5, 0.5 ]
+  test:
+    dataloader:
+      batch_size: 8
+      shuffle: False
+      num_workers: 1
+      pin_memory: False
+      drop_last: False
+    dataset:
+      _type: GenerationUnpairedDatasetWithEdge
+      root_a: "/data/i2i/VoxCeleb2Anime/testA"
+      root_b: "/data/i2i/VoxCeleb2Anime/testB"
+      edges_path: "/data/i2i/VoxCeleb2Anime/edges"
+      edge_type: "hed"
+      random_pair: False
+      size: [128, 128]
+      pipeline:
+        - Load
+        - Resize:
+            size: [128, 128]
+        - ToTensor
+        - Normalize:
+            mean: [ 0.5, 0.5, 0.5 ]
+            std: [ 0.5, 0.5, 0.5 ]
+    video_dataset:
+      _type: SingleFolderDataset
+      root: "/data/i2i/VoxCeleb2Anime/test_video_frames/"
+      with_path: True
+      pipeline:
+        - Load
+        - Resize:
+            size: [ 256, 256 ]
+        - ToTensor
+        - Normalize:
+            mean: [ 0.5, 0.5, 0.5 ]
+            std: [ 0.5, 0.5, 0.5 ]

configs/synthesizers/TAHG.yml

@@ -3,10 +3,6 @@ engine: TAHG
 result_dir: ./result
 max_pairs: 1000000
 
-distributed:
-  model:
-  # broadcast_buffers: False
-
 misc:
   random_seed: 324
 
@@ -23,6 +19,7 @@ interval:
 model:
   generator:
     _type: TAHG-Generator
+    _bn_to_sync_bn: False
     style_in_channels: 3
     content_in_channels: 1
     num_blocks: 4