imx8mq evk Machine learning inference warnings.

[kandarp_rastey]

-Image = LF_v5.10.52-2.1.0_images_IMX8MQEVK

- While trying to run a tflite(object detection) inference on gpu(with NNAPI) i'm facing some strange warnings.

- The same inference runs fine on cpu without any warnings and also much faster than on GPU with warnings.

Warnings go as follows:

INFO: Created TensorFlow Lite delegate for NNAPI.

WARNING: Operator LEAKY_RELU (v1) refused by NNAPI delegate: Unsupported operation type.

WARNING: Operator DEQUANTIZE (v2) refused by NNAPI delegate: kTfLiteFloat16 not supported as input.

WARNING: Operator SPLIT_V (v1) refused by NNAPI delegate: Unsupported operation type.

WARNING: Operator RESIZE_BILINEAR (v3) refused by NNAPI delegate: Operator refused due performance reasons.

Applied NNAPI delegate.
W [query_hardware_caps:66]Unsupported evis version"

- Need help resolving these warnings.

[kandarp_rastey]

Hello @Bio_TICFSL ,

-Will try moving to 5.10.72 BSP and testing the same.

-It'll be great if you can give a timeline or ETA for the release of newer BSP that will have the full NNAPI supported.

Thanks for looking into this though.

[Bio_TICFSL]

Hello kandarp,

NNAPI Some operator are not support such as DEQUATIZE, SPLIT, so you have to move to 5.10.72 BSP or wait for the new release that will have the full NNAPI supported,

Regards 

[kandarp_rastey]

Hello @Bio_TICFSL ,

- Even after Moving to the recommended 5.10.72 BSP the above mentioned issue is still persistent.

- Waiting for the new release that will have the full NNAPI supported ever since ‎"02-10-2022 06:39 AM"

- Expecting a response with the expected date of availability for the latest release that will have the full NNAPI supported.

Regards Kandarp Rastey.