Creating .blob from Google Vertex Edge Model

MikieReed

I am trying to use a model created in Vertex on an OAK. However, I am having major issues using the blob converter web app. I have also tried the openvino toolkits (versions 2021, 2022, and 2023) with no luck. Can you please advise if this is possible?

Here are the possible outputs from Vertex:
https://cloud.google.com/vertex-ai/docs/export/export-edge-model#tf-lite:~:text=Tensorflow.js-,The%20OUTPUT_BUCKET,-you%20specified%20in

Thank you!

erik

Hi MikieReed ,
I would suggest exporting as TFLite, and then using tflite2tensorflow package to convert to OpenVINO. Let us know if you have any issues with that flow.
Thansk, Erik

MikieReed

Thank you, Erik.
I had played around with that a bit, but I will try it again and let you know.
Have a wonderful day!
mikie

MikieReed

Good morning, Erik,
I have been trying the tflite2tensorflow package. Per the package README, I have:

Created a frozed .pb file:

  --model_path 021023_model.tflite \
  --flatc_path ../flatc \
  --schema_path ../schema.fbs \
  --output_pb \
  --optimizing_for_openvino_and_myriad \
  --rigorous_optimization_for_myriad

TensorFlow/Keras model building process complete!
saved_model / .pb output started ====================================================
saved_model / .pb output complete!

However, on the .blob creation step:

tflite2tensorflow \
  --model_path 021023_model.tflite \
  --flatc_path ../flatc \
  --schema_path ../schema.fbs \
  --output_no_quant_float32_tflite \
  --output_dynamic_range_quant_tflite \
  --output_weight_quant_tflite \
  --output_float16_quant_tflite \
  --output_integer_quant_tflite \
  --string_formulas_for_normalization 'data / 255.0' \
  --output_tfjs \
  --output_coreml \
  --output_tftrt_float32 \
  --output_tftrt_float16 \
  --output_onnx \
  --onnx_opset 11 \
  --output_openvino_and_myriad

Myriad Inference Engine blob convertion started ============================================
Traceback (most recent call last):
  File "/usr/local/bin/tflite2tensorflow", line 6608, in <module>
    main()
  File "/usr/local/bin/tflite2tensorflow", line 6578, in main
    shutil.copy(f'{model_output_path}/openvino/FP16/saved_model.xml', f'{model_output_path}/openvino/FP16/saved_model_vino.xml')
  File "/usr/lib/python3.8/shutil.py", line 418, in copy
    copyfile(src, dst, follow_symlinks=follow_symlinks)
  File "/usr/lib/python3.8/shutil.py", line 264, in copyfile
    with open(src, 'rb') as fsrc, open(dst, 'wb') as fdst:
FileNotFoundError: [Errno 2] No such file or directory: 'saved_model/openvino/FP16/saved_model.xml'

I played with it for a few hours, and couldn't find anywhere that that file was supposed to get created.

Please advise. Thank you!

mikie

erik

Hi MikieReed ,
I am not very familiar with the tool, best to submit the issue on their repo. In general, Luxonis officially supports models whose training notebooks can be found at depthai-ml-training. That said, other models can likely still work, but usually need a bit more work to get converted.
Thanks, Erik