Forums - snpe-net-run returns similar latency for quantized and non-quantized dlc

2 posts / 0 new
Last post
snpe-net-run returns similar latency for quantized and non-quantized dlc
pravinair02
Join Date: 17 Jan 19
Posts: 1
Posted: Wed, 2019-01-23 21:47

Hi,

         For MobileNetV1/V2 model, latencies are similar for quantized and non-quantized dlc's when run with snpe-net-run for CPU/GPU/DSP on target. Is it expected?

 

Regards,

P

  • Up0
  • Down0
jihoonk
Profile picture
Join Date: 28 Jan 13
Location: Seoul
Posts: 40
Posted: Wed, 2019-01-23 23:19

Hi pravinair02,

CPU and GPU always use non-quantized model and DSP always use quantized model. So if you provide quantized model in CPU and GPU mode, SNPE automatically dequantize the model in the initialization step and in case of DSP, vice versa. So even if you provide any (quantized or non-quantized) models, the inference time is the same. For details, refer to below link.

https://developer.qualcomm.com/docs/snpe/quantized_models.html

Thanks,

Jihoon

  • Up0
  • Down0
or Register

Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.