[Facenet] Inference Engine Core Dumped

[shieh__joseph]

Hi, Sir

                I have already got the xml and bin by IR stage, my step show as below:

                step 1. set input batch size to [1, 160,160, 3]

                step 2. embeddings value output from network and then do l2_normalize 

                step3. freeze model and feed to mo.py to generate IR xml and bin

                Command:sudo python3 ../model_optimizer/mo_tf.py --input_model ~/tensorflowGPU/tensorflow/facenet_revise.pb --input input --input_shape=[1,160,160,3] --output output --output_dir ir_facenet_revise --data_type FP32 --model_name Facenet

               Step4. when I test inference engine read model -> (set input blob ,output blob) -> infer_request.Infer() -> infer_request.GetBlob(firstOutputName) ------ CoreDumped

                firstOutputName is normalize3251

               Q: why getBlob node name will got the coredumped, I have no idea what to do, need your help or reference to slove it, thanks.

BR,

Joseph

[Zhen_Z_Intel]

Hi joseph,

Would you please try to debug with gdb to see if there's any problem cause by memory over-bounding access, or invalid pointer. Make sure you have enough memory, please check with top, and share the log of your debug info. Thank you.  

Best regards,
Fiona

[shieh__joseph]

Hi, Fiona:

              The gdb debug mode show the error message show as below, have you any suggestion to do next step to clarify this memory issue, many thanks for your helping. my Host PC Hw info show as below:

 Gdb info (Cmake Type=Debug):

[ INFO ] Loading model to the plugin
[New Thread 0x7fffe455f700 (LWP 27694)]
[New Thread 0x7fffe3d5e700 (LWP 27695)]
[New Thread 0x7fffe335c780 (LWP 27696)]
[New Thread 0x7fffe2f5b800 (LWP 27697)]
[New Thread 0x7fffe2b5a880 (LWP 27698)]
[ INFO ] Joseph num_channels is 3
[ INFO ] Joseph image_size is 25600
[ INFO ] Starting inference (1 iterations)

Thread 2 "classification_" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffe455f700 (LWP 27694)]
0x00007fffebdcffba in mkldnn_memory_get_data_handle ()
   from /opt/intel/computer_vision_sdk_2018.1.249/deployment_tools/inference_engine/lib/ubuntu_16.04/intel64/libMKLDNNPlugin.so     

             Attachment is test source code.

             CPU: Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz

             MemTotal:       16345032 kB 

BR,

Joseph

[Jeffrey_M_Intel1]

Sorry for the delayed reply.  Improvements for Facenet will be available in the next release.