Hi,
I benchmarked mobilenet network using Qualcomm SNPE , ARM-NN , and TF-Lite . Results of Arm-NN and TF-lite is comparable but SNPE results are so bad (approx 7x more than Arm-NN/TF-Lite in terms of Latency).
So, I have some questions:
1. Is SNPE supports inter/intra layer multithreading.
2. If yes then Is't tunable and how to enable it ?
3. Is SNPE supports Optimized Ops/kernels (Neon) for operator implementations.
4. One document shared by Qualcomm , says that using SNPE, we can able to achieve 27x gain , so on what metrices , they are comparing. (https://www.qualcomm.com/media/documents/files/achieving-ai-at-scale-pre...)
Thanks and Regards
Amar kumar