Hi.
Is it possible in principle to evaluate NXP eIQ Model Zoo ResNet50 on an FRDM i.MX93 board?
I see that as per NXP eIQ® Model Zoo supported platforms, i.MX93 in general is supported,
but is FRDM i.MX93 supported?
I try to evaluate NXP eIQ Model Zoo ResNet50 performance on an FRDM i.MX93.
I use the FRDM i.MX93 with its default settings: eMMC boot and followed
NXP eIQ Model Zoo ResNet50 README which suggests to create TensorFlow Lite models
by running bash recipe.sh.
To only create (not evaluate) models I needed to overcome the following hurdles:
- downgrade Python from 3.12 to 3.11 via pyenv: TensorFlow 2.12.0 demands 3.11,
the default image provides 3.12; - cross-compile git: pyenv needs it and the default Yocto-based image lacks it;
- create an additional partition on the eMMC: model creation process easily overflows entire root partition with "No space left on device" (this implies boot from SD, gptfdisk compilation and MBR to GPT conversion);
- move eiq-model-zoo repository to this additional partition, create tensorflow_datasets symlink in the home directory.
And all this in vain as I still hit a wall with the kernel OOM Killer.
I can't even create TensorFlow Lite models on the FRDM i.MX93, not to mention test its performance.
Any help on the matter will be appreciated.
