generating TFLITE models for iMX8+

menashesoffer — Tue, 24 Jan 2023 14:27:01 GMT

I am trying to run TFLITE models on a board that contains iMX8+ core (third party board, variscite).
basically I prepare an 8-bit quantized TFLITE model using tensorflow.lite.converter.

the model has at least some validity as I can compile it successfully for a different EDGE device.

however I fail to run it on the iMX8+ (at least not on variscite board).

I have a general question: before asking any specific question, is there any NXP guiding material on how to prepare TFLITE models for iMX8+ (restrictions, unsupported ops, versions and/or flags for the converter, etc.)?
or any post-processing is required for the file that tensorflow.lite.convert() generates?

thanks

Re: generating TFLITE models for iMX8+

Bio_TICFSL — Wed, 25 Jan 2023 14:22:47 GMT

Hello,

You can check the eIQ blog from:

https://community.nxp.com/t5/Blogs/eIQ-Sample-Apps-TFLite-Quantization/ba-p/1131121

Regards

topic generating TFLITE models for iMX8+ in i.MX Processors

generating TFLITE models for iMX8+

Re: generating TFLITE models for iMX8+