Hello all!
Mostly the forums have been very helpful and after a few iterations of playing swapping faces I am getting the hang of it. However, I am wanting to try to pull more detail out and thus would like to try using Gblock and Fully connected, split layers.
I am getting errors when trying to create models with either or both of these enabled.
I have reinstalled and checked all of my versions; the installer installed Tensorflow 2.10.1 . It could not install cudatoolkit 11.2 or cudnn8.1 via pip or conda, so I have manually installed these.
(it might be helpful to let people know the maximum supported versions? There seems some confusion as to which versions of which will work etc.)
I'm running an rtx3070 with 8gb of ram. Driver version is 565.90 with cuda version 12.7.
The model is Phaze-A, I'm using clipv 16-64 with an input size of 256. I was scaling it to 114 encoder scaling but that caused errors in loading weights. I still get the error if I don't load the weights at all.
The error I get when I enable Gblock is
Code: Select all
03/05/2025 19:10:42 MainProcess MainThread settings __init__ INFO Enabling Mixed Precision Training.
03/05/2025 19:10:42 MainProcess MainThread settings _get_strategy DEBUG Using strategy: <tensorflow.python.distribute.distribute_lib._DefaultDistributionStrategy object at 0x000001D1E78665F0>
03/05/2025 19:10:42 MainProcess MainThread settings __init__ DEBUG Initialized Settings
03/05/2025 19:10:42 MainProcess MainThread settings __init__ DEBUG Initializing Loss: (color_order: bgr)
03/05/2025 19:10:42 MainProcess MainThread settings _get_mask_channels DEBUG uses_masks: (True, True, True), mask_channels: [3, 4, 5]
03/05/2025 19:10:42 MainProcess MainThread settings __init__ DEBUG Initialized: Loss
03/05/2025 19:10:42 MainProcess MainThread model __init__ DEBUG Initialized ModelBase (Model)
03/05/2025 19:10:42 MainProcess MainThread phaze_a _get_input_shape DEBUG Encoder input set to: (224, 224, 3)
03/05/2025 19:10:42 MainProcess MainThread phaze_a build DEBUG New model, inference or summary. Falling back to default build: (exists: False, inference: False, is_summary: True)
03/05/2025 19:10:42 MainProcess MainThread settings strategy_scope DEBUG Using strategy scope: <tensorflow.python.distribute.distribute_lib._DefaultDistributionContext object at 0x000001D1E751BE00>
03/05/2025 19:10:42 MainProcess MainThread model _get_inputs DEBUG Getting inputs
03/05/2025 19:10:42 MainProcess MainThread model _get_inputs DEBUG inputs: [<KerasTensor: shape=(None, 224, 224, 3) dtype=float32 (created by layer 'face_in_a')>, <KerasTensor: shape=(None, 224, 224, 3) dtype=float32 (created by layer 'face_in_b')>]
03/05/2025 19:10:42 MainProcess MainThread phaze_a __call__ DEBUG Scaling to (0, 1) for 'clipv_farl-b-16-64'
03/05/2025 19:10:42 MainProcess MainThread clip __init__ DEBUG Initializing: ViT (name: FaRL-B-16-64, input_size: 224, load_weights: True)
03/05/2025 19:10:42 MainProcess MainThread clip __init__ DEBUG Initializing: VisualTransformer (input_resolution: 224, patch_size: 16, width: 768, layers: 12, heads: 12, output_dim: 512, name: visual)
03/05/2025 19:10:42 MainProcess MainThread clip __init__ DEBUG Initialized: VisualTransformer
03/05/2025 19:10:42 MainProcess MainThread clip __init__ DEBUG Initialized: ViT
03/05/2025 19:10:42 MainProcess MainThread clip __init__ DEBUG Initializing: Transformer (width: 768, num_layers: 12, heads: 12, attn_mask: None, name: visual.transformer)
03/05/2025 19:10:42 MainProcess MainThread clip __init__ DEBUG Initialized: Transformer
03/05/2025 19:10:42 MainProcess MainThread clip __call__ DEBUG Calling Transformer with input: (None, 197, 768)
03/05/2025 19:10:42 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.0
03/05/2025 19:10:42 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.1
03/05/2025 19:10:42 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.2
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.3
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.4
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.5
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.6
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.7
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.8
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.9
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.10
03/05/2025 19:10:43 MainProcess MainThread clip _get_name DEBUG Generating block name: visual.transformer.resblocks.11
03/05/2025 19:10:43 MainProcess MainThread utils _get DEBUG Model exists: D:\faceswap\.fs_cache\CLIPv_FaRL-B-16-64_v1.h5
03/05/2025 19:10:43 MainProcess MainThread clip __call__ INFO Loading CLIPv trained weights for 'FaRL-B-16-64'
03/05/2025 19:10:43 MainProcess MainThread attrs __getitem__ DEBUG Creating converter from 3 to 5
03/05/2025 19:10:45 MainProcess MainThread phaze_a _build_encoders DEBUG Encoders: {'a': <KerasTensor: shape=(None, 512) dtype=float16 (created by layer 'encoder')>, 'b': <KerasTensor: shape=(None, 512) dtype=float16 (created by layer 'encoder')>}
03/05/2025 19:10:45 MainProcess MainThread phaze_a __init__ DEBUG Initializing: FullyConnected (side: both, input_shape: (512,))
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:10:45 MainProcess MainThread phaze_a __init__ DEBUG Initialized: FullyConnected (side: both, min_nodes: 512, max_nodes: 512)
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:10:45 MainProcess MainThread phaze_a _get_curve DEBUG Obtaining curve: (start_y: 512, end_y: 512, num_points: 1, scale: 0.0, mode: full)
03/05/2025 19:10:45 MainProcess MainThread phaze_a _get_curve DEBUG Returning curve: [512]
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 128, scaled_filters: 128
03/05/2025 19:10:45 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_128_0
03/05/2025 19:10:45 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_0. filters: 128, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:10:45 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_0_conv2d, filters: 512, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:10:45 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x000001D209771C30>)
03/05/2025 19:10:45 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x000001D209772200>
03/05/2025 19:10:45 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 512, 128]
03/05/2025 19:10:45 MainProcess MainThread initializers _create_basis DEBUG filters_size: 128, filters: 512, size: 6, dtype: float32
03/05/2025 19:10:45 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (512, 6, 6, 128), Output shape: (512, 3, 3, 512)
03/05/2025 19:10:45 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 512, 512)
03/05/2025 19:10:45 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_128_1
03/05/2025 19:10:45 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_1. filters: 128, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:10:45 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_1_conv2d, filters: 512, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:10:45 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x000001D2097B6800>)
03/05/2025 19:10:45 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x000001D2097B57B0>
03/05/2025 19:10:45 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 128]
03/05/2025 19:10:45 MainProcess MainThread initializers _create_basis DEBUG filters_size: 128, filters: 128, size: 6, dtype: float32
03/05/2025 19:10:45 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 128), Output shape: (128, 3, 3, 512)
03/05/2025 19:10:45 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 512)
03/05/2025 19:10:45 MainProcess MainThread phaze_a __init__ DEBUG Initializing: UpscaleBlocks (side: both, layer_indicies: (0, 1))
03/05/2025 19:10:45 MainProcess MainThread phaze_a __init__ DEBUG Initialized: UpscaleBlocks
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 4, new_dim: 4
03/05/2025 19:10:45 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 4, new_dim: 4
03/05/2025 19:10:45 MainProcess MainThread phaze_a _get_curve DEBUG Obtaining curve: (start_y: 512, end_y: 16, num_points: 6, scale: 0.5, mode: cap_max)
03/05/2025 19:10:45 MainProcess MainThread phaze_a _get_curve DEBUG Returning curve: [512, 256, 128, 64, 32, 16]
03/05/2025 19:10:45 MainProcess MainThread phaze_a __call__ DEBUG Generated class filters: [512, 256, 128, 64, 32, 16]
03/05/2025 19:10:45 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_512_0
03/05/2025 19:10:46 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_0. filters: 512, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:10:46 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_0_conv2d, filters: 2048, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:10:46 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x000001D2097B5510>)
03/05/2025 19:10:46 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x000001D2097B4D60>
03/05/2025 19:10:46 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 512]
03/05/2025 19:10:46 MainProcess MainThread initializers _create_basis DEBUG filters_size: 512, filters: 128, size: 6, dtype: float32
03/05/2025 19:10:46 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 512), Output shape: (128, 3, 3, 2048)
03/05/2025 19:10:46 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 2048)
03/05/2025 19:10:46 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_512_1
03/05/2025 19:10:46 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_1. filters: 512, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:10:46 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_1_conv2d, filters: 2048, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:10:46 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x000001D2097B69B0>)
03/05/2025 19:10:46 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x000001D20979E2F0>
03/05/2025 19:10:46 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 512]
03/05/2025 19:10:46 MainProcess MainThread initializers _create_basis DEBUG filters_size: 512, filters: 128, size: 6, dtype: float32
03/05/2025 19:10:46 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 512), Output shape: (128, 3, 3, 2048)
03/05/2025 19:10:46 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 2048)
03/05/2025 19:10:46 MainProcess MainThread phaze_a __init__ DEBUG Initializing: FullyConnected (side: gblock, input_shape: (512,))
03/05/2025 19:10:46 MainProcess MainThread phaze_a __init__ DEBUG Initialized: FullyConnected (side: gblock, min_nodes: 512, max_nodes: 512)
03/05/2025 19:10:46 MainProcess MainThread phaze_a _get_curve DEBUG Obtaining curve: (start_y: 512, end_y: 512, num_points: 3, scale: -0.5, mode: full)
03/05/2025 19:10:46 MainProcess MainThread phaze_a _get_curve DEBUG Returning curve: [512, 512, 512]
03/05/2025 19:10:46 MainProcess MainThread phaze_a _build_fully_connected DEBUG Fully Connected: {'a': [[<KerasTensor: shape=(None, 8, 8, 512) dtype=float16 (created by layer 'fc_both')>, <KerasTensor: shape=(None, 8, 8, 512) dtype=float16 (created by layer 'fc_both')>], <KerasTensor: shape=(None, 512) dtype=float16 (created by layer 'fc_gblock')>], 'b': [[<KerasTensor: shape=(None, 8, 8, 512) dtype=float16 (created by layer 'fc_both')>, <KerasTensor: shape=(None, 8, 8, 512) dtype=float16 (created by layer 'fc_both')>], <KerasTensor: shape=(None, 512) dtype=float16 (created by layer 'fc_gblock')>]}
Traceback (most recent call last):
File "D:\faceswap\lib\cli\launcher.py", line 225, in execute_script
process.process()
File "D:\faceswap\scripts\train.py", line 197, in process
self._load_model()
File "D:\faceswap\scripts\train.py", line 285, in _load_model
model.build()
File "D:\faceswap\plugins\train\model\phaze_a.py", line 186, in build
super().build()
File "D:\faceswap\plugins\train\model\_base\model.py", line 271, in build
self._model = self.build_model(inputs)
File "D:\faceswap\plugins\train\model\phaze_a.py", line 339, in build_model
g_blocks = self._build_g_blocks(inters)
File "D:\faceswap\plugins\train\model\phaze_a.py", line 435, in _build_g_blocks
input_shapes = [K.int_shape(inter)[1:] for inter in inputs["a"]]
File "D:\faceswap\plugins\train\model\phaze_a.py", line 435, in <listcomp>
input_shapes = [K.int_shape(inter)[1:] for inter in inputs["a"]]
File "C:\Users\admin\MiniConda3\envs\faceswap\lib\site-packages\keras\backend.py", line 1531, in int_shape
shape = x.shape
AttributeError: 'list' object has no attribute 'shape' When I enable fully connected layers I get
Code: Select all
03/05/2025 19:14:19 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:19 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:14:19 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:19 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 128, scaled_filters: 128
03/05/2025 19:14:19 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_128_2
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_2. filters: 128, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_2_conv2d, filters: 512, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:19 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EAFB700>)
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EAFA110>
03/05/2025 19:14:19 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 512, 128]
03/05/2025 19:14:19 MainProcess MainThread initializers _create_basis DEBUG filters_size: 128, filters: 512, size: 6, dtype: float32
03/05/2025 19:14:19 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (512, 6, 6, 128), Output shape: (512, 3, 3, 512)
03/05/2025 19:14:19 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 512, 512)
03/05/2025 19:14:19 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_128_3
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_3. filters: 128, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_3_conv2d, filters: 512, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:19 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EAFA5C0>)
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EAFA0E0>
03/05/2025 19:14:19 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 128]
03/05/2025 19:14:19 MainProcess MainThread initializers _create_basis DEBUG filters_size: 128, filters: 128, size: 6, dtype: float32
03/05/2025 19:14:19 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 128), Output shape: (128, 3, 3, 512)
03/05/2025 19:14:19 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 512)
03/05/2025 19:14:19 MainProcess MainThread phaze_a __init__ DEBUG Initializing: UpscaleBlocks (side: b, layer_indicies: (0, 1))
03/05/2025 19:14:19 MainProcess MainThread phaze_a __init__ DEBUG Initialized: UpscaleBlocks
03/05/2025 19:14:19 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 4, new_dim: 4
03/05/2025 19:14:19 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 4, new_dim: 4
03/05/2025 19:14:19 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_512_2
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_2. filters: 512, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_2_conv2d, filters: 2048, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:19 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EABFD30>)
03/05/2025 19:14:19 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EABCF40>
03/05/2025 19:14:19 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 512]
03/05/2025 19:14:19 MainProcess MainThread initializers _create_basis DEBUG filters_size: 512, filters: 128, size: 6, dtype: float32
03/05/2025 19:14:20 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 512), Output shape: (128, 3, 3, 2048)
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 2048)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_512_3
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_3. filters: 512, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_3_conv2d, filters: 2048, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EABCC70>)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EABE500>
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 512]
03/05/2025 19:14:20 MainProcess MainThread initializers _create_basis DEBUG filters_size: 512, filters: 128, size: 6, dtype: float32
03/05/2025 19:14:20 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 512), Output shape: (128, 3, 3, 2048)
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 2048)
03/05/2025 19:14:20 MainProcess MainThread phaze_a __init__ DEBUG Initializing: FullyConnected (side: shared, input_shape: (512,))
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:14:20 MainProcess MainThread phaze_a __init__ DEBUG Initialized: FullyConnected (side: shared, min_nodes: 512, max_nodes: 512)
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:14:20 MainProcess MainThread phaze_a _get_curve DEBUG Obtaining curve: (start_y: 512, end_y: 512, num_points: 1, scale: 0.0, mode: full)
03/05/2025 19:14:20 MainProcess MainThread phaze_a _get_curve DEBUG Returning curve: [512]
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 512, scaled_filters: 512
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 3, new_dim: 2
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_filters DEBUG original_filters: 128, scaled_filters: 128
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_128_4
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_4. filters: 128, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_4_conv2d, filters: 512, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EAFB280>)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EAFA650>
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 512, 128]
03/05/2025 19:14:20 MainProcess MainThread initializers _create_basis DEBUG filters_size: 128, filters: 512, size: 6, dtype: float32
03/05/2025 19:14:20 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (512, 6, 6, 128), Output shape: (512, 3, 3, 512)
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 512, 512)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_128_5
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_5. filters: 128, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_128_5_conv2d, filters: 512, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EABDE10>)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EA20CA0>
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 128]
03/05/2025 19:14:20 MainProcess MainThread initializers _create_basis DEBUG filters_size: 128, filters: 128, size: 6, dtype: float32
03/05/2025 19:14:20 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 128), Output shape: (128, 3, 3, 512)
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 512)
03/05/2025 19:14:20 MainProcess MainThread phaze_a __init__ DEBUG Initializing: UpscaleBlocks (side: shared, layer_indicies: (0, 1))
03/05/2025 19:14:20 MainProcess MainThread phaze_a __init__ DEBUG Initialized: UpscaleBlocks
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 4, new_dim: 4
03/05/2025 19:14:20 MainProcess MainThread phaze_a _scale_dim DEBUG target_resolution: 256, original_dim: 4, new_dim: 4
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_512_4
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_4. filters: 512, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_4_conv2d, filters: 2048, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EA4CD90>)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EA4FB20>
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 512]
03/05/2025 19:14:20 MainProcess MainThread initializers _create_basis DEBUG filters_size: 512, filters: 128, size: 6, dtype: float32
03/05/2025 19:14:20 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 512), Output shape: (128, 3, 3, 2048)
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 2048)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_name DEBUG Generating block name: upscale_512_5
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_5. filters: 512, kernel_size: 3, padding: same, scale_factor: 2, normalization: None, activation: leakyrelu, kwargs: {})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG name: upscale_512_5_conv2d, filters: 2048, kernel_size: 3, strides: (1, 1), padding: same, normalization: None, activation: leakyrelu, use_depthwise: False, kwargs: {'is_upscale': True})
03/05/2025 19:14:20 MainProcess MainThread nn_blocks _get_default_initializer DEBUG Set default kernel_initializer: (original: None current: <lib.model.initializers.ConvolutionAware object at 0x0000028F6EA4E500>)
03/05/2025 19:14:20 MainProcess MainThread nn_blocks __init__ DEBUG Using ICNR Initializer: <lib.model.initializers.ICNR object at 0x0000028F6EA65930>
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ INFO Calculating Convolution Aware Initializer for shape: [3, 3, 128, 512]
03/05/2025 19:14:20 MainProcess MainThread initializers _create_basis DEBUG filters_size: 512, filters: 128, size: 6, dtype: float32
03/05/2025 19:14:20 MainProcess MainThread initializers _space_to_depth DEBUG Input shape: (128, 6, 6, 512), Output shape: (128, 3, 3, 2048)
03/05/2025 19:14:20 MainProcess MainThread initializers __call__ DEBUG Output shape: (3, 3, 128, 2048)
Traceback (most recent call last):
File "D:\faceswap\lib\cli\launcher.py", line 225, in execute_script
process.process()
File "D:\faceswap\scripts\train.py", line 197, in process
self._load_model()
File "D:\faceswap\scripts\train.py", line 285, in _load_model
model.build()
File "D:\faceswap\plugins\train\model\phaze_a.py", line 186, in build
super().build()
File "D:\faceswap\plugins\train\model\_base\model.py", line 271, in build
self._model = self.build_model(inputs)
File "D:\faceswap\plugins\train\model\phaze_a.py", line 338, in build_model
inters = self._build_fully_connected(encoders)
File "D:\faceswap\plugins\train\model\phaze_a.py", line 399, in _build_fully_connected
inter_a = [kl.Concatenate(name="inter_a")([inter_a[0], fc_shared(inputs["a"])])]
File "C:\Users\admin\MiniConda3\envs\faceswap\lib\site-packages\keras\utils\traceback_utils.py", line 70, in error_handler
raise e.with_traceback(filtered_tb) from None
File "C:\Users\admin\MiniConda3\envs\faceswap\lib\site-packages\keras\layers\merging\concatenate.py", line 98, in build
raise ValueError(
ValueError: A `Concatenate` layer should be called on a list of at least 1 input. Received: input_shape=[[(None, 8, 8, 512), (None, 8, 8, 512)], [(None, 8, 8, 512), (None, 8, 8, 512)]] I have no issues running the model at all with these options disabled, I can run it up to batch size 7 with about 12k images each side. I have tried instantiating the model with a batch size of one, still doesn't work, also disabling other things like conv aware or icr init makes no difference. I'm happy to paste the entire model summary but it's huge of course, let me know if this will help. I'm assuming its some kind of error with Keras? But I have no idea. (My Keras version is 3.9.0 incidentally). I've tried other encoders and I still get the errors. I have no clue how to interpret this!
Thanks in advance!