TDA4VM: YOLOV2,3,5 deploy on sk-tda4vm

Hi,

Currently we are supporting YoloX with different flavours available  'yolox_s_lite', 'yolox_tiny_lite', 'yolox_nano_lite', 'yolox_pico_lite', 'yolox_femto_lite'.

We currently support YoloV5 out of other listed versions.

If you are trying out of the box example models, it might not be offloaded to full acceleration on C7x-MMA, few non supported kernal from TIDL will run on ARM core.

Regards,

Pratik

Hi,

TIDL offload will happen for layers which are supported as part of TIDL optimized kernals.

The non supported layers will get offloaded to arm core (Non Optimized flow)

So for any network which has all the operators/kernals supported as part of TIDL will run in optimized way.

We recommend to check supported operator list here : https://github.com/TexasInstruments/edgeai-tidl-tools/blob/master/docs/supported_ops_rts_versions.md and compare with any model you are trying to infer.

Regards,

Pratik

As mentioned,

For any Neural Network you are experimenting with

If all the operators/kernal are supported by TIDL library, compiled network graph will get offloaded to C7x-MMA (In layman terms, It will run in optimized mode).

You can check the list of supported operators here : 

Pratik Kedar said:

We recommend to check supported operator list here : https://github.com/TexasInstruments/edgeai-tidl-tools/blob/master/docs/supported_ops_rts_versions.md and compare with any model you are trying to infer.

Regards,

Pratik