model optimizer converting tensorflow slim fine-tuned inception v3

[cfu__cfu]

Hi,

I fine tuned the tensorflow slim inception v3 image classification model by running the script 

https://github.com/tensorflow/models/blob/master/research/slim/scripts/finetune_inception_v3_on_flowers.sh

Then, I try to freeze the graph by running

python3 /tensorflow/tensorflow/python/tools/freeze_graph.py --input_graph=/tmp/flowers-models/inception_v3/graph.pbtxt  --input_checkpoint=/tmp/flowers-models/inception_v3/model.ckpt-1000 --input_binary=false --output_graph=/tmp/frozen_inception_v3.pb --output_node_names=InceptionV3/Predictions/Reshape_1

and got the ifle

/tmp/frozen_inception_v3.pb

Then I try to convert the model by running

python3 mo_tf.py --input_model /tmp/frozen_inception_v3.pb -b 1 --mean_value [127.5,127.5,127.5] --scale 127.5

and got the error:

Model Optimizer arguments:
Common parameters:
    - Path to the Input Model:     /data/train/frozen_inception_v3.pb
    - Path for generated IR:     /opt/intel/openvino_2019.1.094/deployment_tools/model_optimizer/.
    - IR output name:     frozen_inception_v3
    - Log level:     ERROR
    - Batch:     1
    - Input layers:     Not specified, inherited from the model
    - Output layers:     Not specified, inherited from the model
    - Input shapes:     Not specified, inherited from the model
    - Mean values:     [127.5,127.5,127.5]
    - Scale values:     Not specified
    - Scale factor:     127.5
    - Precision of IR:     FP32
    - Enable fusing:     True
    - Enable grouped convolutions fusing:     True
    - Move mean values to preprocess section:     False
    - Reverse input channels:     False
TensorFlow specific parameters:
    - Input model in text protobuf format:     False
    - Path to model dump for TensorBoard:     None
    - List of shared libraries with TensorFlow custom layers implementation:     None
    - Update the configuration file with input/output node names:     None
    - Use configuration file used to generate the model with Object Detection API:     None
    - Operations to offload:     None
    - Patterns to offload:     None
    - Use the config file:     None
Model Optimizer version:     2019.1.0-341-gc9b66a2
WARNING: Logging before flag parsing goes to stderr.
E0414 10:12:34.278545 140149041551104 main.py:317] Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.front.input_cut.InputCut'>): Graph contains 0 node after executing <class 'extensions.front.input_cut.InputCut'>. It considered as error because resulting IR will be empty which is not usual

any suggestion? 

[SuryaPSC_Intel]

Hi cfu,

As your model is a Tensorflow*- Slim Image classification model.

Please follow the steps at Converting TensorFlow*-Slim Image Classification Model.

Best Regards,

Surya

[cfu__cfu]

Hi Chauhan,

I could convert the pre-trained slim image classification model without any issue.

However, after I fine-tune and freeze the model, I am not able to convert it to IR. 

[SuryaPSC_Intel]

Hi cfu,

Did you generate the inference graph for inception v3 using the following command(for inception v1) given in the documentation.

python3 tf_models/research/slim/export_inference_graph.py \
    --model_name inception_v1 \
    --output_file inception_v1_inference_graph.pb

Did you try passing the ckpt file to the model optimizer as given in the documentation?

<MODEL_OPTIMIZER_INSTALL_DIR>/mo_tf.py --input_model ./inception_v1_inference_graph.pb --input_checkpoint ./inception_v1.ckpt -b 1 --mean_value [127.5,127.5,127.5] --scale 127.5

Also, please share the output you get on passing the model to summarize_graph.py

Best Regards,

Surya

[cfu__cfu]

Hi Chauhan,

Thanks for the reply, instead of fine-tune inception v3, I try to start with inception v1 first.

I follow the script https://github.com/tensorflow/models/blob/master/research/slim/scripts/finetune_inception_v1_on_flowers.sh

to fine-tune a pre-retrained model, you can find the files under folder fine-tune through below link:

https://drive.google.com/drive/folders/1zFJwYAZSj-aII4wbDqpoQq8_L0ynV6m0?usp=sharing ;

Then, I freeze the model by running

python3 tensorflow/python/tools/freeze_graph.py --input_graph graph.pbtxt --input_checkpoint model.ckpt-3000 --output_graph /tmp/inception_v1_freeze.pb --output_node_names InceptionV1/Logits/Predictions/Softmax

Then, I try to convert the freezed model to IR by running

python3 mo_tf.py --input_model /tmp/inception_v1_freeze.pb -b 1 --mean_value [127.5,127.5,127.5] --scale 127.5

but got the error, the tensorflow version is 1.15.2

Model Optimizer arguments:
Common parameters:
	- Path to the Input Model: 	/tmp/inception_v1_freeze.pb
	- Path for generated IR: 	/opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/.
	- IR output name: 	inception_v1_freeze
	- Log level: 	ERROR
	- Batch: 	1
	- Input layers: 	Not specified, inherited from the model
	- Output layers: 	Not specified, inherited from the model
	- Input shapes: 	Not specified, inherited from the model
	- Mean values: 	[127.5,127.5,127.5]
	- Scale values: 	Not specified
	- Scale factor: 	127.5
	- Precision of IR: 	FP32
	- Enable fusing: 	True
	- Enable grouped convolutions fusing: 	True
	- Move mean values to preprocess section: 	False
	- Reverse input channels: 	False
TensorFlow specific parameters:
	- Input model in text protobuf format: 	False
	- Path to model dump for TensorBoard: 	None
	- List of shared libraries with TensorFlow custom layers implementation: 	None
	- Update the configuration file with input/output node names: 	None
	- Use configuration file used to generate the model with Object Detection API: 	None
	- Use the config file: 	None
Model Optimizer version: 	2020.2.0-60-g0bc66e26ff
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.front.output_cut.OutputCut'>): Graph contains 0 node after executing <class 'extensions.front.output_cut.OutputCut'>. It considered as error because resulting IR will be empty which is not usual

[SuryaPSC_Intel]

Hi cfu,

Can you please attached the ckpt file also in the above mentioned drive link. So, that we may replicate at our end.

Best Regards,

Surya

[cfu__cfu]

Hi Chauhan,

I have attached the ckpt file inception_v1.ckpt 

[SuryaPSC_Intel]

Hi cfu,

Please use the following command to generate IR. 

sudo python3 mo.py --input_model /home/surya/Downloads/inception_v1_inference_graph.pb --input_checkpoint /home/surya/Downloads/inception_v1.ckpt -b 1 --mean_value [127.5,127.5,127.5] --scale 127.5
Model Optimizer arguments:
Common parameters:
	- Path to the Input Model: 	/home/surya/Downloads/inception_v1_inference_graph.pb
	- Path for generated IR: 	/opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/.
	- IR output name: 	inception_v1_inference_graph
	- Log level: 	ERROR
	- Batch: 	1
	- Input layers: 	Not specified, inherited from the model
	- Output layers: 	Not specified, inherited from the model
	- Input shapes: 	Not specified, inherited from the model
	- Mean values: 	[127.5,127.5,127.5]
	- Scale values: 	Not specified
	- Scale factor: 	127.5
	- Precision of IR: 	FP32
	- Enable fusing: 	True
	- Enable grouped convolutions fusing: 	True
	- Move mean values to preprocess section: 	False
	- Reverse input channels: 	False
TensorFlow specific parameters:
	- Input model in text protobuf format: 	False
	- Path to model dump for TensorBoard: 	None
	- List of shared libraries with TensorFlow custom layers implementation: 	None
	- Update the configuration file with input/output node names: 	None
	- Use configuration file used to generate the model with Object Detection API: 	None
	- Use the config file: 	None
Model Optimizer version: 	2020.2.0-60-g0bc66e26ff

[ SUCCESS ] Generated IR version 10 model.
[ SUCCESS ] XML file: /opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/./inception_v1_inference_graph.xml
[ SUCCESS ] BIN file: /opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/./inception_v1_inference_graph.bin
[ SUCCESS ] Total execution time: 13.22 seconds. 
[ SUCCESS ] Memory consumed: 491 MB.

[cfu__cfu]

Hi Chauhan,

I can produce the IR model with inception_v1.ckpt. However, I am not able to convert the fine-tune model to IR.  

[SuryaPSC_Intel]

Hi cfu,

Can you please confirm you are generating the inference graph for inception v3 using:

python3 tf_models/research/slim/export_inference_graph.py \
    --model_name inception_v3 \
    --output_file inception_v3_inference_graph.pb

Please share the output on passing the inference_graph to summarize_graph.py and model optimizer.

Best regards,

Surya