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!

[Error 24] Toolkit Error: >1 image inference not supported.

Hi guys I have created a Caffe version of my IDC Classifier:

https://github.com/BreastCancerAI/IDC-Classifier/blob/master/Caffe/CaffeNet/

There is one difference between the prototxt files online and the one I use and that is the last layer (fc8) is set to num_output: 2 swapped from 1000 to represent the 2 classes. When I run this through mvNCCompile I receive the following error:

mvNCCompile deploy.prototxt -o caffeGraph -w model_iter_5000.caffemodel

[Error 24] Toolkit Error: >1 image inference not supported.

Any suggestions on how to resolve this ?

Comments

  • 4 Comments sorted by Votes Date Added
  • @Tome_at_Intel hi mate sorry to tag, just wondered if you had any input on this only found it mentioned twice on Google and one is an unresolved post on here :( Any suggestions? I ruled out the issue of the device being faulty as I can still compile a tensorflow graph. Hoping to get past this :) Thanks in advance mate.

  • @AdamMiltonBarker Hi. No problem at all. I believe you can solve this issue by changing the input dimensions. The NCSDK only supports a batch size of 1 and by default this network has a batch size of 10. You can see it in the input param of the data layer at the very top of the network. The first dimension is the batch size, the second is the channel size and the remaining two are the input resolution h and w.

    So this is the original:

    name: "CaffeNet"
    layer {
      name: "data"
      type: "Input"
      top: "data"
      input_param { shape: { dim: 10 dim: 3 dim: 227 dim: 227 } }
    }
    

    Change the batch size to 1 and you should have:

    name: "CaffeNet"
    layer {
      name: "data"
      type: "Input"
      top: "data"
      input_param { shape: { dim: 1 dim: 3 dim: 227 dim: 227 } }
    }
    
  • Perfect thank you I thought it may be my prototxt files, thanks will update and retrain now.

  • @AdamMiltonBarker I don't think you need to retrain your network to make this change unless it is for a unrelated reason. You can just change the batch size in the prototxt file.

Sign In or Register to comment.