Keras to tf.keras

Has anyone ported this (awesome) tutorial to tf.keras from keras? My motivation is related to training in AWS SageMaker where there is some required functionality in a tf.contrib package.

For the most part this seems pretty straight forward but I hit a wall (exceeded my talent) on keras_ssd300.py

  File "/home/ubuntu/projects/rev-tensorflow-ssd/models/keras_ssd300.py", line 318, in ssd_300
    conv4_3_norm = L2Normalization(gamma_init=20, name='conv4_3_norm')(conv4_3)
  File "/home/ubuntu/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 538, in __call__
    self._maybe_build(inputs)
  File "/home/ubuntu/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 1603, in _maybe_build
    self.build(input_shapes)
  File "/home/ubuntu/projects/rev-tensorflow-ssd/models/keras_layers/keras_layer_L2Normalization.py", line 64, in build
    self.trainable_weights = [self.gamma]
  File "/home/ubuntu/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 1639, in __setattr__
    super(Layer, self).__setattr__(name, value)
AttributeError: can't set attribute

If I understand this correctly, this is feeding conv4_3 into the L2 normalization layer It is trying to set the attribute: self.trainable_weights = [self.gamma] from a(n added) print statement, it looks like self.gamma = [<tf.Variable ‘conv4_3_norm/conv4_3_norm_gamma:0’ shape=(512,) dtype=float32>]

Can anyone offer a clue here?

[MaAmini]

tried the same fix and got

“ValueError: Dimensions must be equal, but are 512 and 64 for ‘conv4_3_norm/mul’ (op: ‘Mul’) with input shapes: [?,64,64,512], [64].” tf 1.13

I also got the same error. Do you find any solution for that?

[Maioy97]

tried the same fix and got

“ValueError: Dimensions must be equal, but are 512 and 64 for ‘conv4_3_norm/mul’ (op: ‘Mul’) with input shapes: [?,64,64,512], [64].” tf 1.13