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

Custom Resnet-50 is not converted to movidius model properly

I tried to transfer my custom resnet-50 ( https://github.com/cvjena/cnn-models) to movidius graph but I failed.

 

I got some message when I transfer caffe model to movidius model and I got nan value when I tried to inference my model using NCSDK and converted graph file.

 

root@hyunwoo-desk:/ncappzoo/apps/book# mvNCCompile -s 12 resnet50_deploy.prototxt -w resnet50_custom_transfer_v2_iter_180000.caffemodel -o 0629_model.graph

 

mvNCCompile v02.00, Copyright @ Intel Corporation 2017

 

/usr/local/bin/ncsdk/Controllers/FileIO.py:65: UserWarning: You are using a large type. Consider reducing your data sizes for best performance

 

Blob generated

 

root@hyunwoo-desk:/ncappzoo/apps/book# ./run_flowai.sh

 

[ nan nan nan nan nan]

 

[ nan nan nan nan nan]

 

[ nan nan nan nan nan]

 

[ nan nan nan nan nan]

 

[ nan nan nan nan nan]

 

[ nan nan nan nan nan]
Tags (1)
0 Kudos
5 Replies
idata
Community Manager
63 Views

@hyunwoolee I am getting a hang when converting this model using the mvNCCompile tool. Still investigating the issue. Thank you for your patience.

idata
Community Manager
63 Views

@Tome_at_Intel Hi. any update on this thread?

idata
Community Manager
63 Views

@namsulee It looks like the device is being lost, but I'm still not quite sure why this is happening.

 

Just to make sure I am using the right weights, are you using the weights from the link https://github.com/cvjena/cnn-models/blob/master/ResNet_preact/ResNet50_cvgj/model_download_link.txt?

idata
Community Manager
63 Views

@Tome_at_Intel yes, I am using that , I did transfer learning based on that model.

idata
Community Manager
63 Views

@hyunwoolee Please try using your model with NCSDK v 2.08 and let me know if your issue has been resolved. https://github.com/movidius/ncsdk/tree/ncsdk2