23333

configs/synthesizers/GauGAN.yml

@@ -0,0 +1,167 @@
+name: huawei-GauGAN-3
+engine: GauGAN
+result_dir: ./result
+max_pairs: 1000000
+
+misc:
+  random_seed: 324
+
+handler:
+  clear_cuda_cache: True
+  set_epoch_for_dist_sampler: True
+  checkpoint:
+    epoch_interval: 1 # checkpoint once per `epoch_interval` epoch
+    n_saved: 2
+  tensorboard:
+    scalar: 100 # log scalar `scalar` times per epoch
+    image: 4 # log image `image` times per epoch
+  test:
+    random: True
+    images: 10
+
+model:
+  generator:
+    _type: SPADEGenerator
+    _add_spectral_norm: True
+    in_channels: 3
+    out_channels: 3
+    num_blocks: 7
+    use_vae: False
+    num_z_dim: 256
+#  discriminator:
+#    _type: MultiScaleDiscriminator
+#    _add_spectral_norm: True
+#    num_scale: 2
+#    down_sample_method: "bilinear"
+#    discriminator_cfg:
+#      _type: PatchDiscriminator
+#      in_channels: 3
+#      base_channels: 64
+#      num_conv: 4
+#      need_intermediate_feature: True
+  discriminator:
+    _type: PatchDiscriminator
+    _add_spectral_norm: True
+    in_channels: 3
+    base_channels: 64
+    num_conv: 4
+    need_intermediate_feature: True
+
+
+loss:
+  gan:
+    loss_type: hinge
+    weight: 1.0
+    real_label_val: 1
+    fake_label_val: 0.0
+  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: 2
+  mgc:
+    weight: 5
+  fm:
+    weight: 5
+  edge:
+    weight: 0
+    hed_pretrained_model_path: ./network-bsds500.pytorch
+
+optimizers:
+  generator:
+    _type: Adam
+    lr: 1e-4
+    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: 1
+      shuffle: True
+      num_workers: 2
+      pin_memory: True
+      drop_last: True
+    dataset:
+      _type: GenerationUnpairedDataset
+      root_a: "/data/face2cartoon/all_face"
+      root_b: "/data/selfie2anime/trainB/"
+      random_pair: True
+      pipeline_a:
+        - Load
+        - RandomCrop:
+            size: [ 178, 178 ]
+        - Resize:
+            size: [ 256, 256 ]
+        - RandomHorizontalFlip
+        - ToTensor
+        - Normalize:
+            mean: [ 0.5, 0.5, 0.5 ]
+            std: [ 0.5, 0.5, 0.5 ]
+      pipeline_b:
+        - Load
+        - Resize:
+            size: [ 286, 286 ]
+        - RandomCrop:
+            size: [ 256, 256 ]
+        - RandomHorizontalFlip
+        - ToTensor
+        - Normalize:
+            mean: [ 0.5, 0.5, 0.5 ]
+            std: [ 0.5, 0.5, 0.5 ]
+  test:
+    which: video_dataset
+    dataloader:
+      batch_size: 1
+      shuffle: False
+      num_workers: 1
+      pin_memory: False
+      drop_last: False
+    dataset:
+      _type: GenerationUnpairedDataset
+      root_a: "/data/face2cartoon/test/human"
+      root_b: "/data/face2cartoon/test/anime"
+      random_pair: True
+      pipeline_a:
+        - Load
+        - Resize:
+            size: [ 256, 256 ]
+        - ToTensor
+        - Normalize:
+            mean: [ 0.5, 0.5, 0.5 ]
+            std: [ 0.5, 0.5, 0.5 ]
+      pipeline_b:
+        - Load
+        - Resize:
+            size: [ 256, 256 ]
+        - 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 ]