TypeError: Tensor objects are not iterable when eager execution is not enabled. To iterate over this tensor use tf.map_fn.

[mosheliv]

Hi,
Trying to optimize pretty standard resnet50 model from keras and getting this:

Traceback (most recent call last):
  File "rn50_inf.py", line 289, in <module>
    model = reduce_keras_model(model)
  File "/home/m/.local/lib/python2.7/site-packages/kito/__init__.py", line 371, in reduce_keras_model
    for f in prev_layer:
  File "/home/m/.local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 431, in __iter__
    "Tensor objects are not iterable when eager execution is not "
TypeError: Tensor objects are not iterable when eager execution is not enabled. To iterate over this tensor use tf.map_fn.

keras 2.2.4
tf 1.10.0

[ZFTurbo]

Can you please run this script in your environment?
https://github.com/ZFTurbo/Keras-inference-time-optimizer/blob/master/kito/test_bench.py

[mosheliv]

attached
test_bench.txt

[ZFTurbo]

And now the same but change line
models_to_test = ['mobilenet_small']
to
models_to_test = ['resnet50']

[mosheliv]

attached
test_bench.txt

[ZFTurbo]

ResNet50 from keras was successfully optimized on your machine. You probably have some error in your model. Can you print model.summary() for your implementation?

[mosheliv]

I will do that tomorrow as it is almost 1am... Thank you!

…

On Wed, Dec 19, 2018, 00:35 ZFTurbo ***@***.*** wrote: ResNet50 from keras was successfully optimized on your machine. You probably have some error in your model. Can you print model.summary() for your implementation? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL6LCwfIBRlXa6V2LRT6KWVMUMlrAvC9ks5u6NLwgaJpZM4ZX8oi> .

[mosheliv]

I think I know what the problem is. One of the issues state that it doesn't handle nested models well.

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         (None, 512, 512, 3)       0         
_________________________________________________________________
resnet50 (Model)             (None, 2048)              23587712  
_________________________________________________________________
dense_1 (Dense)              (None, 28)                57372     
=================================================================
Total params: 23,645,084
Trainable params: 23,591,964
Non-trainable params: 53,120

[ZFTurbo]

I actually had such case in test_bench.py: 'multi_model_layer' and it works fine. I will check your case and probably fix the code soon.

[mosheliv]

let me know if you need any other information from me. this is the code that creates the model: def create_model(input_shape, n_out): inp_mask = Input(shape=input_shape) pretrain_model_mask = ResNet50(input_shape = (512,512,3), include_top=False, weights='imagenet', pooling='avg') pretrain_model_mask.name='densenet_mask' x = pretrain_model_mask(inp_mask) out = Dense(n_out, activation='sigmoid')(x) model = Model(inputs=inp_mask, outputs=[out]) return model

…

On Thu, Dec 20, 2018 at 8:55 AM ZFTurbo ***@***.***> wrote: I actually had such case in test_bench.py: 'multi_model_layer' and it works fine. I will check your case and probably fix the code soon. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL6LC4mP5LZl5rPym0T80t3LCL5ZkRqVks5u6pmygaJpZM4ZX8oi> .

[ZFTurbo]

I fixed the problem. Your code should work fine now.

[mosheliv]

I was just looking sadly at densenet201 doing inference! Thank you!

…

On Fri, Dec 21, 2018, 12:22 ZFTurbo ***@***.*** wrote: I fixed the problem. Your code should work fine now. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL6LC9bAjNIjoavzjU-pwPxJKazMbIh0ks5u7BvFgaJpZM4ZX8oi> .

[linjianxia1031]

I have the same problem, could you tell me how to fix it?