BlazeFace back_256x256_float32 vs back_256x256_integer_quant #457
Description
Issue Type
Support
OS
Windows
OS architecture
x86_64
Programming Language
Python
Framework
TensorFlowLite
Model name and Weights/Checkpoints URL
https://github.com/PINTO0309/PINTO_model_zoo/tree/main/030_BlazeFace
Description
face_detection_back_256x256_float32.tflite
face_detection_back_256x256_integer_quant.tflite
hello,
Before moving the model to an embedded platform, I try to run it in python environment.
I made a script that run face_detection_back_256x256_float32.tflite and it works great.
when I try to switch the model to face_detection_back_256x256_integer_quant.tflite I get vastly different results although the IO is in the same format.
Ofc there will be a difference between the float and the int versions, but I wonder if there is another preprocess/ postprocess step that need to be added when using the integer_quant model (again, it is NOT full quant, it takes float and outputs float)
Thank you
Relevant Log Output
URL or source code for simple inference testing code
No response