Community
cancel
Showing results for 
Search instead for 
Did you mean: 
idata
Community Manager
380 Views

Inference result is changed after transforted to movidius model

Hi, I'm hyunwoo lee

 

I'm using the movidius and raspberry pi to recognize classify images.

 

But after I convert caffe model to movidius graph , I got the different result.

 

So I checked same image for two models ( one is caffe model , the other is movidius model)

 

I used Resnet-18 model and my target class is 5.

 

As you can see , result is so different .

 

CAFFE / resnet18_deploy / resnet18_custom_transfer_iter_5000

 

[0.15182055201006436, 0.15381028087897108, 0.3939189286761392, 0.15019196295911535, 0.15025827547570988]

 

CAFFE / resnet18_deploy / resnet18_custom_transfer_v2_iter_4000

 

[0.2008975630376957, 0.1603913024819096, 0.19200574522089883, 0.16627960206785145, 0.2804257871916444]

 

MOVIDIUS / resnet18_custom_transfer_iter_5000.graph

 

[ 4.99725342e-03 2.81494141e-01 7.12890625e-01 0.00000000e+00 5.64575195e-04]

 

MOVIDIUS / resnet18_custom_transfer_v2_iter_4000.graph

 

[ 9.86328125e-01 8.58306885e-04 5.98144531e-03 5.03921509e-03 1.46579742e-03]
0 Kudos
3 Replies
idata
Community Manager
67 Views

@hyunwoolee Hi, how did you check your ResNet model on the NCS? Can you provide your NCS code and ResNet 18 model for testing? With the NCS, you may have to perform some manual image preprocessing (mean subtraction and scaling) before inference. WIth Caffe, the preprocessing is done automatically.

idata
Community Manager
67 Views

idata
Community Manager
67 Views

@Tome_at_Intel Yes, That's my model . I also checked preprocessing but I didn't used mean subtraction and scaling while inferencing model using caffe and movidius .