Closed ysh329 closed 5 years ago
I don't think ncnn has support nnlib yet. For Mace and SNPE, they have almost the same performance. For model prepare time, Mace is better.
Excuse, I'm little confused, what's the nnlib? Is an Android common API for deep learning? Does it have raw implementation?
It's a library which can be found in https://developer.qualcomm.com/software/hexagon-dsp-sdk as hexagon nn, you can get all the information from Hexagon DSP SDK. So basically, it's for Hexagon DSP and is usually integrated with deep learning inference engines, e.g., SNPE, MACE. The benchmark results of DSP of MACE and SNPE all use nnlib.
Greatly thank you, bin.
Besides, I wanna ask what's the relationship between DSP/AIP runtimes of SNPE and Hexagon NN. Of course, I wanna know the performance between them. Any idea?
DSP of SNPE is based on Hexagon NN, so its performance is also based on Hexagon NN.
Greatly and always Thank you ~~ bin
Hi, all!
Has any benchmark result for nnlib with mace, ncnn or etc? Besides, I wanna ask which performance is better between nnlib and SNPE?
Thanks in advance.