TypeError: Exception encountered when calling layer "custom_crop_layer_2" " f"(type CustomCropLayer). #63788
Comments
|
@pzy2000, |
tilakrayal
added
comp:keras
sorry, I forgot to share the model file, it is here: |
google-ml-butler
bot
removed
the
stat:awaiting response
|
@pzy2000, And also this issue is related to Keras, Could you please reach out to keras-team/keras repo for the quick response. Thank you! |
tilakrayal
added
the
stat:awaiting response
|
This issue is stale because it has been open for 7 days with no activity. It will be closed if no further activity occurs. Thank you. |
github-actions
bot
added
the
stale
|
This issue was closed because it has been inactive for 7 days since being marked as stale. Please reopen if you'd like to work on this further. |
Issue type
Bug
Have you reproduced the bug with TensorFlow Nightly?
Yes
Source
binary
TensorFlow version
tf 2.10.0
Custom code
Yes
OS platform and distribution
Linux Ubuntu 20.04.4
Mobile device
No response
Python version
3.7.6
Bazel version
No response
GCC/compiler version
No response
CUDA/cuDNN version
12.2
GPU model and memory
No response
Current behavior?
Expect:
successfully saved to tf format ckpt.
Now:
Traceback (most recent call last):
File "issue_tf2onnx_001.py", line 207, in
model.save(f'tmp/lenet5_sb', save_format='tf')
File "/opt/conda/envs/tf2onnx/lib/python3.7/site-packages/keras/utils/traceback_utils.py", line 70, in error_handler
raise e.with_traceback(filtered_tb) from None
File "/opt/conda/envs/tf2onnx/lib/python3.7/contextlib.py", line 119, in exit
next(self.gen)
File "issue_tf2onnx_001.py", line 77, in call
indices.append(slice(0 + crop[0], shape - crop[1]))
TypeError: Exception encountered when calling layer "custom_crop_layer_2" " f"(type CustomCropLayer).
unsupported operand type(s) for -: 'NoneType' and 'int'
Call arguments received by layer "custom_crop_layer_2" " f"(type CustomCropLayer):
• args=('tf.Tensor(shape=(None, None, None, 32), dtype=float32)',)
• kwargs={'training': 'False'}
Standalone code to reproduce the issue
import keras import tensorflow as tf class CustomCastLayer(keras.layers.Layer): def __init__(self, target_dtype, **kwargs): self.target_dtype = target_dtype super().__init__(**kwargs) def build(self, input_shape): """ Do Nothing """ super().build(input_shape) def call(self, inputs, **kwargs): output = tf.cast(inputs, self.target_dtype) return output def get_config(self): config = super().get_config() config.update({"target_dtype": self.target_dtype}) return config def compute_output_shape(self, input_shape): """ Do Nothing """ return input_shape class CustomPadLayer(keras.layers.Layer): def __init__(self, padding=[[0, 0]], constant_values=0, **kwargs): self.padding = padding # add [0,0] to padding so we will not change the shape of batch dimension self.constant_values = constant_values super().__init__(**kwargs) def build(self, input_shape): super().build(input_shape) def call(self, inputs, **kwargs): output = tf.pad(inputs, paddings=tf.constant([[0, 0]] + self.padding), mode="CONSTANT", constant_values=self.constant_values) return output def get_config(self): config = super().get_config() config.update({"padding": self.padding, "constant_values": self.constant_values}) return config def compute_output_shape(self, input_shape): """ Formula to calculate the output shape Suppose the input_shape is (None, N, W, C), the `paddings` is [[a, b], [c, d]]. The output_shape is (None, N+a+b, W+c+d, C). """ input_shape_list = list(input_shape) padding = [[0, 0]] + self.padding assert len(input_shape_list) == len(padding) # Two dimensions should match. output_shape = [None] for i, pad in zip(input_shape_list[1:], padding[1:]): output_shape.append(i + pad[0] + pad[1]) return tuple(output_shape) class CustomCropLayer(keras.layers.Layer): def __init__(self, cropping, **kwargs): self.cropping = cropping super().__init__(**kwargs) def build(self, input_shape): super().build(input_shape) def call(self, inputs, **kwargs): input_shape = inputs.shape.as_list() indices = [slice(None)] cropping = [[0, 0]] + self.cropping # add [0,0] to padding so we will not change the shape of batch dimension for shape, crop in zip(input_shape[1:], cropping[1:]): indices.append(slice(0 + crop[0], shape - crop[1])) return inputs[indices] def get_config(self): config = super().get_config() config.update({"cropping": self.cropping}) return config def compute_output_shape(self, input_shape): """ Formula to calculate the output shape Suppose the input_shape is (None, N, W, C), the `cropping` is [[a, b], [c, d]]. The output_shape is (None, N-a-b, W-c-d, C). """ input_shape_list = list(input_shape) cropping = [[0, 0]] + self.cropping assert len(input_shape_list) == len(cropping) # Two dimensions should match. output_shape = [None] for i, crop in zip(input_shape[1:], cropping[1:]): output_shape.append(i - crop[0] - crop[1]) return tuple(output_shape) class CustomExpandLayer(keras.layers.Layer): def __init__(self, axis=1, **kwargs): self.axis = axis super().__init__(**kwargs) def build(self, input_shape): super().build(input_shape) def call(self, inputs, **kwargs): return tf.expand_dims(inputs, self.axis) def get_config(self): config = super().get_config() config.update({"axis": self.axis}) return config def compute_output_shape(self, input_shape): """ Formula to calculate the output shape Suppose the input_shape is [None, N, W, C]: axis=0: output_shape: [1, None, N, W, C] axis=1 (default): output_shape: [None, 1, N, W, C] axis=2: output_shape: [None, N, 1, W, C] axis=3: output_shape: [None, N, W, 1, C] axis=4: output_shape: [None, N, W, C, 1] axis=5: raise Exception """ input_shape_list = list(input_shape) if self.axis > len(input_shape_list): raise ValueError(f"axis {self.axis} should be smaller than input_shape + 1: {len(input_shape_list) + 1}") output_shape = input_shape_list[0:self.axis] + [1] + input_shape_list[self.axis:] return tuple(output_shape) # we should use tuple!!! not list !!! class CustomDropDimLayer(keras.layers.Layer): def __init__(self, axis=1, **kwargs): self.axis = axis super().__init__(**kwargs) def build(self, input_shape): super().build(input_shape) def call(self, inputs, **kwargs): """ Something magic to automatically generate indices for array slicing. To determine a specific axis, we can use slice(None) to replace `:` """ dim = len(inputs.shape) if self.axis > dim - 1 or self.axis < 1: raise ValueError(f"axis: {self.axis} should be within the range: [1, {dim - 1}] for {dim}D tensor") indices = [slice(None) for i in range(dim)] indices[self.axis] = 0 return inputs[indices] def get_config(self): config = super().get_config() config.update({"axis": self.axis}) return config def compute_output_shape(self, input_shape): """ Formula to calculate the output shape Suppose the input_shape is [None, N, W, C]: axis=0: # Although it is feasible, we don't allow this to happen Raise Exception axis=1 (default): output_shape: [None, W, C] axis=2: output_shape: [None, N, C] axis=3: output_shape: [None, N, W] axis=4: Raise Exception """ input_shape_list = list(input_shape) output_shape = input_shape_list[0:self.axis] + input_shape_list[self.axis + 1:] return tuple(output_shape) def custom_objects(mode="custom"): def no_activation(x): return x def leakyrelu(x): import keras.backend as K return K.relu(x, alpha=0.01) # objects = {} objects = {'no_activation': no_activation, 'leakyrelu': leakyrelu} if mode == "custom": objects['CustomPadLayer'] = CustomPadLayer objects['CustomCropLayer'] = CustomCropLayer objects['CustomDropDimLayer'] = CustomDropDimLayer objects['CustomExpandLayer'] = CustomExpandLayer objects['CustomCastLayer'] = CustomCastLayer return objects model_path = "/data1/pzy/MUTANTS/LEMON/lenet5_fashion/lenet5-fashion-mnist_origin-NLAll30-LMerg38-Edge63-NLAll90-92/lenet5-fashion-mnist_origin-NLAll30-LMerg38-Edge63-NLAll90.h5" model = tf.keras.models.load_model(model_path, custom_objects=custom_objects()) # model = tf.keras.models.load_model(model_path,) model.save(f'tmp/lenet5_sb', save_format='tf')Relevant log output
The text was updated successfully, but these errors were encountered: