Hi ,
I'm running RensNet-50 v1 model on a Samsung Galaxy S9+ device usind SNPE (1.19.2). I'm noticing that the GPU (fp32 and fp16) performance of caffe model is about 2X better than tensoflow.
Is this because the layers are falling back to CPU more on tensorflow model when compared to caffe ?
Can someone be able to explain why this might be happening.
Thanks
Manasa
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.
Hi,
You can confirm the network layers supported by NPE SDK for Caffe and TensorFlow frameworks here
By using the Benchmarking tool provided by the NPE SDK, you can know the inference time of individual network layer in architecture and how they are performing on different run times. You can find the documentation for Benchmarking tool here.
Hope this helps you in analyzing the inference time in detail and compare it for different frameworks.