Edgeai TIDL Tools and Examples - This repository contains Tools and example developed for Deep learning runtime (DLRT) offering provided by TI’s edge AI solutions.
I tested my onnx pose model inference speed on the TDA4VM board, which has generated artifacts files using edgeai-tidl-tools tools.
I found that the inference speed of the CPU and DSP on the board was similar, about 1.4 seconds, but other models I tested would have about 70 to 90 times the inference speed of the CPU in the DSP.
I observed the differences between my pose model and other models and found that my model uses 3D convolution, and other models use 2D convolution. Is my model infering slow because of 3D convolution or not?
BTW, I want to know, which operators are supported by the DSP of the TDA4VM board now? And does the DSP support transformer operator calculation?
Hi
I tested my onnx pose model inference speed on the TDA4VM board, which has generated artifacts files using edgeai-tidl-tools tools.
I found that the inference speed of the CPU and DSP on the board was similar, about 1.4 seconds, but other models I tested would have about 70 to 90 times the inference speed of the CPU in the DSP.
I observed the differences between my pose model and other models and found that my model uses 3D convolution, and other models use 2D convolution. Is my model infering slow because of 3D convolution or not?
BTW, I want to know, which operators are supported by the DSP of the TDA4VM board now? And does the DSP support transformer operator calculation?
And my model is shown as below:
Thanks!!!