Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6570 Discussões

Model Optimizer TensorFlow: Failed to run transform 'create_ir'

Julien1
Principiante
938 Visualizações

Computer Vision SDK 2017.1.163
Ubuntu 16.04

I have successfully converted an Inception V1 TensorFlow model using Model Optimizer. I am now trying to convert a customized squeezenet tensorflow model. Summarize graph output for the frozen graph is as follows:

bazel-bin/tensorflow/tools/graph_transforms/summarize_graph --in_graph=out-256-192.pb

Found 1 possible inputs: (name=input, type=float(1), shape=[1,192,256,3])
No variables spotted.
Found 1 possible outputs: (name=output, op=Cast)
Found 1646509 (1.65M) const parameters, 0 (0) variable parameters, and 0 control_edges
54 nodes assigned to device '/device:CPU:0'Op types used: 112 Const, 64 Identity, 32 BiasAdd, 30 Conv2D, 28 Relu, 10 StridedSlice, 8 ConcatV2, 5 Mul, 4 MaxPool, 3 Pack, 2 Add, 2 Conv2DBackpropInput, 2 Cast, 1 Placeholder, 1 ResizeBilinear, 1 Shape, 1 Sigmoid, 1 Sub
To use with tensorflow/tools/benchmark:benchmark_model try these arguments:
bazel run tensorflow/tools/benchmark:benchmark_model -- --graph=out-256-192.pb --show_flops --input_layer=input --input_layer_type=float --input_layer_shape=1,192,256,3 --output_layer=output

My command line for Model Optimizer was:

python3 /opt/intel/computer_vision_sdk_2017.1.163/mo/model_optimizer_tensorflow/modeloptimizer/scripts/model_optimizer.py --input_model=out-256-192.pb --input=input --output=output --data_type=float32 --input_shape=1,192,256,3 --model_name intel-model

...
Validating IR graph...
IR graph validation FAILED for the following nodes
    id: 367 node_name: inference/conv2d_transpose node_type: op type: Conv2DBackpropInput
    id: 369 node_name: inference/conv2d_transpose:0 node_type: tensor type:
    id: 370 node_name: inference/BiasAdd_2 node_type: op type: BiasAdd
    id: 371 node_name: inference/BiasAdd_2:0 node_type: tensor type:
    id: 374 node_name: inference/Shape_2 node_type: op type: Shape
    id: 375 node_name: inference/Shape_2:0 node_type: tensor type:
    id: 386 node_name: inference/stack_1 node_type: op type: Pack
    id: 388 node_name: inference/stack_1:0 node_type: tensor type:
    id: 389 node_name: inference/conv2d_transpose_1 node_type: op type: Conv2DBackpropInput
    id: 390 node_name: inference/conv2d_transpose_1:0 node_type: tensor type:
    id: 391 node_name: inference/BiasAdd_4 node_type: op type: BiasAdd
    id: 392 node_name: inference/BiasAdd_4:0 node_type: tensor type:
    id: 401 node_name: inference/ResizeBilinear node_type: op type: ResizeBilinear
    id: 403 node_name: inference/ResizeBilinear:0 node_type: tensor type:
    id: 408 node_name: result node_type: op type: Cast
    id: 409 node_name: result:0 node_type: tensor type:
    id: 410 node_name: output node_type: op type: Cast
    id: 411 node_name: output:0 node_type: tensor type:
Something went wrong. Please check the error and try again:
<class 'Exception'>
Failed to run transform 'create_ir'
WARNING: Failed to run the following transform(s): create_ir

Any advice appreciated.

Thanks,

Julien.

 

0 Kudos
1 Solução
Anna_B_Intel
Funcionário
938 Visualizações

Hi Julien, 

Currently SqueezeNet is unsupported by Model Optimizer for TensorFlow. You have 2 options: use SqueezeNet from Caffe (it's supported by Model Optimizer for Caffe) or wait for the next release of Computer Vision SDK where SqueezeNet in Tensorflow framework will be supported.

Best wishes, 

Anna 

 

Ver solução na publicação original

2 Respostas
Anna_B_Intel
Funcionário
939 Visualizações

Hi Julien, 

Currently SqueezeNet is unsupported by Model Optimizer for TensorFlow. You have 2 options: use SqueezeNet from Caffe (it's supported by Model Optimizer for Caffe) or wait for the next release of Computer Vision SDK where SqueezeNet in Tensorflow framework will be supported.

Best wishes, 

Anna 

 

Julien1
Principiante
938 Visualizações

Hi Anna - thanks again for your quick reply. I will try the caffe route.

Julien.

Responder