TensorFlow ERROR in Model Optimizer Converter

[Stefano_M_]

Hello,

I trained a MobileNet V2 with TensorFlow on Ubuntu 16.04 following this guide:

1) https://becominghuman.ai/tensorflow-object-detection-api-tutorial-training-and-evaluating-custom-object-detector-ed2594afcf73

2) after the end of the train phase I converted the model as described here: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/exporting_models.md

Now I'm trying, with model optimizer the model:

sudo python3 mo_tf.py --input_model /home/vigilate/TestTrain/frozen_inference_graph.pb --input_checkpoint /home/vigilate/TestTrain/checkpoint

but I'm getting this error:

[ ERROR ]  Cannot load input model: Unable to open table file /home/vigilate/TestTrain/checkpoint: Data loss: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?

I tried also to not use the 2) conversion and run mo_tf.py in this way:

sudo python3 mo_tf.py --input_meta_graph /home/vigilate/training_mobnetv2_07_09_2018/model.ckpt-200.meta

I didn't get any error but the obtained files seem empty:

generated .xml

<?xml version="1.0" ?>
<net batch="1" name="model.ckpt-200" version="2">
    <layers/>
    <edges/>
</net>

generated .bin [dimension 0 Bytes]

I attach here the file TestTrain.zip where I put the file obtained after 2)

Thanks

Stefano

[Mark_L_Intel1]

Hi Stefano,

You might get wrong arguments based on following document:

https://software.intel.com/en-us/articles/OpenVINO-Using-TensorFlow

I tried the data from your zip file and I got succeed, could you double check?

cd TestTrain
python3 ~/deployment_tools/model_optimizer/mo_tf.py --input_model=frozen_inference_graph.pb --tensorflow_use_custom_operations_config ~l/deployment_tools/model_optimizer/extensions/front/tf/ssd_v2_support.json --tensorflow_object_detection_api_pipeline_config pipeline.config --output_dir .

Mark

[Stefano_M_]

Thank you for you reply.

I got succeeded.

Only one thing. I tried the same procedure on Windows, but the conversion with model optimizer fails.

The same conversion (same files, same input parameters) works on Ubuntu