search

google-coral / project-bodypix

BodyPix model demo application for Google Coral
Apache License 2.0
275 stars 52 forks source link

TF lite version #14

Closed dvk17are closed 3 years ago

dvk17are commented 3 years ago

Hi, thanks for your great work guys :) It would be nice if you also release the tflite models from the previous stage. I am currently working on decoding the output without BasicEngine and I would like to test it on CPU only with the tflite interpreter.

With the TPU version, tensors are not allocated due to custom operations: interpreter= Interpreter(model_path='bodypix_resnet_50_640_480_16_quant_edgetpu_decoder.tflite') interpreter.allocate_tensors()

Error : RuntimeError: Encountered unresolved custom op: edgetpu-custom-op.Node number 0 (edgetpu-custom-op) failed to prepare. Encountered unresolved custom op: edgetpu-custom-op.Node number 0 (edgetpu-custom-op) failed to prepare.

Namburger commented 3 years ago

@dvk17are hello, you'll need to register the posenet_decoder.so like how I do it here: https://github.com/google-coral/project-posenet/blob/ugrade-to-frogfish/pose_engine.py#L79-L83

Naveen-Dodda commented 3 years ago

@dvk17are answering your first question are you looking for cpu based bodypix models a step before edgetpu_compiler ?

if so they are here. Model without _edgetpu.tfllite extension belongs to cpu.

Naveen-Dodda commented 3 years ago

I am closing this issue, as the request has been fulfilled. Please feel free to reopen if you have more questions