Hi,
- My customer wants to check if the processing time of TIDL_OD is reasonable or not, considering TDA2x performance.
- They prepared two object detection models from caffe-jacinto model, Initial model and Sparse model.
- Input data size: 768 x 320
- Run these models on TDA2x, using TIDL_OD usecase, and then checked ALG_TIDL processing time (Local Link Latency avg)
Followings are the test result:
(1) Initial model: ALG_TIDL processing time
EVE = 360 ms, DSP = 17 ms
(2) Sparse model: ALG_TIDL processing time
EVE = 182 ms, DSP = 17 ms
Are these values reasonable one or not?
FYI, I attach configuration files, prototxt, and logs.
Thanks & Regards,
-Shibata