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!

Does the toolkit support caffe model with image as output? And debug issue.

Hi. I have noticed that the sample models were all caffe models for image classification, with "prob" as the output. I tried to launch my model of stereo matching on NCS but it doesn't work and give me the following traceback without any other error message:

File "../../src/./mvNCCheck.py", line 117, in
File "../../src/./mvNCCheck.py", line 91, in check_net
File "../../src/./Controllers/CaffeParser.py", line 919, in parse_caffe
KeyError: 'data'

I wonder whether the toolkit support models with images as output (like stereo matching / image enhancement / super resolution)?

Whether it support models with images of more than 3 channels as input (like a 16 bit depth image)?

Plus, is there any way to see the code for CaffeParser.py so that I can debug? Now I can only see the pyc file but not the original py file.

Comments

  • 2 Comments sorted by Votes Date Added
  • @incredibleup The NCS SDK should have no issue with models using image outputs or models that use more than 3 channels. At the current moment, we are not planning to release the source code for CaffeParser.py. If you are still experiencing problems, please feel free to supply a link to the network in question, so that the issue may be further investigated. Thanks.

  • Hi @Tome_at_Intel,

    Is there an example of a model using image outputs? Ideally Tensorflow based.

    Thanks..

Sign In or Register to comment.