Wrong classification result using the app example in DDK

Greeting to everyone,

When I ran the example in DDK on the Hikey 970, the prediction was wrong. Do anybody meet the same issue?

I just replace the hiai.cambricon with the ResNet_50.cambricon I created using the tool in DDK, and change the input dimension to match the input of ResNet_50 in the classify_jni.cpp. But the prediction was wrong, for example, it predicted cat with ruler. So, how can I fix it.

Board based on the Kirin 970 - HI3670 Application Processor
More info: http://www.96boards.org/product/hikey970/ (Website coming soon…)
Buy now: https://www.seeedstudio.com/HiKey-970-Development-Board-p-3046.html

To help you one would need to able to reproduce the issue you are seing.
Can you specify which example (directory/name/…) in which version of the DDK (link) you are using - on which version of which operating system?

Did you use a pre-trained model or did you train the model yourself?

if you just change input dimension on an existing model it will not function correctly anymore, because it is not trained for your new type input.


@Yifan_Zeng @OJG: Can you explain how did you manage to make the code work on Hikey 970 ? I was trying to run sample demo code given in DDK. But it was not working. It showed me ‘load libhiai.so fail’ and ‘model is incompatible. please run it on CPU’

Did you flash Linux on 970 or used Android for running code ? Were you using Android studio over Android Wifi ADB? I just couldn’t find out any way to sort it out. Please help me.