Hi,
I used the snpe-dlc-quantize tools to quantize a Caffe model MobilenetV1 and ran it on dsp,result was same as cpu.But when I quantized the same Tensorflow model ,dsp result was wrong.I know quantization may decrease result accuracy,but what's the difference between caffe and tf?
thanks!
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 lvxiangxiang,
DSP takes only quantized model in SNPE, and Non-quantized are used by CPU/GPU. If you are using Quantized model for CPU/GPU it will dequantize the model. Similarly vice versa for the DSP. According to the Caffe and Tensorflow, They have their own layers implementation in the framework, so because of that result may vary.
thank you for your reply!
So is there any way to solve the problem that quantized tf mobilenet model has really low accuracy?And,have anyone tried it and result?