OpenVino 2018.3 Model Optimizer raises error while generating IR for TensorFlow MobileNet v1 and v2

[GUPTA__ANMOL]

Hi,

I am unable to generate IR for MobileNet v1 and v2 Networks (In TensorFlow) implemented using Model Optimizer

I picked up the topologies present in the link:

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

On the Same Link I followed the instructions specified in the section "Loading Non-Frozen Models to the Model Optimizer"

Since tar model folder had .meta files (and no checkpoint file), I followed the command 

mo_tf.py --input_meta_graph <INPUT_META_GRAPH>.meta

On running the command, I started getting following errors:

Model Optimizer version: 1.2.185.5335e231

[ ERROR ]  Cannot load input model: Unsuccessful TensorSliceReader constructor: Failed to find any matching files for mobilenet_v1_1.0_224

[[Node: save/RestoreV2_366 = RestoreV2[dtypes=[DT_FLOAT], _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_save/Const_0_0, save/RestoreV2_366/tensor_names, save/RestoreV2_366/shape_and_slices)]]

Thanks

[Monique_J_Intel]

Hi Anmol,

Regarding mobilenet_v1_1.0_224,

I reproduced your issue and this is a bug that we will fix. In the meantime, you can run the following and convert the model:

mo_tf.py --input_model mobilenet_v1_1.0_224_frozen.pb --input_shape [1,224,224,3]

Regarding mobilenet_v2_1.0_224,

I reproduced your issue and this is a bug as well as there is an issue interpreting the 'InfeedEnqueueTuple' node before successfully freezing the model. I also tried to convert the frozen .pb file and ran into an error where the input shapes cannot be inferred from node "MobilenetV2/Conv/Conv2D". The work around for this model is to download freeze_graph.py from here. Run the following command to freeze the .pbtxt file:

python3 freeze_graph.py --input_graph mobilenet_v2_1.0_224_eval.pbtxt --input_checkpoint mobilenet_v2_1.0_224.ckpt --binary=false --output_node_names=MobilenetV2/Predictions/Reshape_1 --output_graph mobilenet_v2_frozen.pb

Now, you have your frozen mobilenet v2 model and can convert it running this:

python3 mo.py --input_model mobilenet_v2_frozen.pb --input_shape [1,224,224,3]

Kind Regards,

Monique Jones

[Kolla__Tarun]

Hey, I kind of have the same problem. I have a unfrozen model and I'm running the following command:

python3 mo_tf.py --input_meta_graph /Users/tarunkolla/Desktop/model/model.ckpt.meta

Following the documentation to load unfrozen model from https://docs.openvinotoolkit.org/2019_R1/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html#freeze-the-tensorflow-model

I get the following error:

[ ERROR ]  The directory "/opt/intel/openvino_2019.1.090/deployment_tools/model_optimizer/." is not writable

Could you help. Thank you.

[Shubha_R_Intel]

Dear Kolla, Tarun,

You either need to open up the shell from which you instantiate the Model Optimizer python script as Administrator (or sudo) or you need to use the --output_dir switch to dump the generated IR into a directory where you have write permissions. 

Thanks,

Shubha