frame

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Any documetation or sugestion to load new caffe or tensorflow model into NCS?

I have installed the SDK successfully,and run the examples in the SDK successfully.Now,I want to run a new caffe model within the NCS.And I typed " mvNCProfile lenet.prototxt -w lenet_iter_10000.caffemodel -s 12 -in input -o lenet.graph ".However it gives me the error of "[Error 24] Toolkit Error: >1 image inference not supported."
So,I am wondering about is there any documentations to tell new users about how to bring new caffe or tensorflow model into NCS?

Poll not found

    Comments

    • 3 Comments sorted by Votes Date Added
    • I am with you. I have seen Youtube examples on how to install the apps and run make files but
      nothing on the end to end implementation.
      Steps are Run build and run graph in tensorflow export from
      tensorflow an inception.meta file and Inception.graph file
      Then use those to run mvNCCompile
      mvNCCompile inception-v1.meta -s 12 -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -is 224 224 -o InceptionV1.graph

      You can then run mvNCProfile to see the preformance
      and run
      mvNCCheck network.meta [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename] [-i image filename] [-id Top-1 Validation ID] [-S scale factor] [-M Mean Subtraction Number or npy filename] [-cs Color Sequence]
      mvNCCheck inception_v1.meta -s 12 -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -is 224 224 -o InceptionV1.graph -cs 0,1,2

      to do the actual inference.. So I think

      And why do they use the word "network " when everyone else uses GRAPH?
      And why do I always have to compile? Cant I precompile load and run?
      I am STILL very new to all of this so anything I mention can be WAY off base.

    • @chicagobob123,thanks for your reply.So you mean that when I get a new tensorflow model, I need to do it with mvNCCompile firstly,and then do it with mvNCCheck.And if I want to see the performance and all layers,I can do it with mvNCProfile,right?
      And another question is that which parameter can be omitted,such as -o,-is,-i,-id and -M,when is needed and when can be omitted?

    • I got my information from here https://movidius.github.io/ncsdk/
      I am on the run today but this should give you an answers.

    Sign In or Register to comment.