2

We have access to a trained tensor-flow CNN model in .tflite format.

We need to convert it into a Linux executable that (trusted) users can call with say, an image as input, and get an object location as output from the model.

This is because we do not want users to be able to trivially have access to the trained model parameters (for legal reasons) - so the plan is to give only executable privilege to the binary, and no read/write access.

Any suggestions ? Would this be possible in something like PyInstaller ?

Thanks !!

Indra Gopal
  • 21
  • 1
  • This is a fantasy, it will be possible to reverse engineer everything without much effort. – Dr. Snoopy Jul 03 '22 at 00:19
  • Yeah! It is possible to reverse engineer the executable files. But if you want to really hide all the details of the model , You can deploy it to the GCP/AWS instances and ask the trusted users to communicate through the get/post method through API keys. –  Jul 07 '22 at 12:04
  • @TensorflowSupport How do you reverse-engineer the .exe without read access to the binary, but only executable access ? In other words, is there a difference between exposing the model through API keys vs. exposing the model through giving it __x privileges for the group ? – Indra Gopal Jul 13 '22 at 12:57
  • @IndraGopal! That was said keeping point of view in advanced users. Pyinstaller is one of solution mentioned solution in this [thread](https://stackoverflow.com/questions/5458048/how-can-i-make-a-python-script-standalone-executable-to-run-without-any-dependen). I suggested API keys as they are more secured compared to binary files. Thank you! –  Jul 14 '22 at 01:54

0 Answers0