Pretrained U-NET Parse Error in Digits 6.0.0

[nigelparsad]

I am trying to run the U-NET semantic segmentation pretrained model on some test data to get a feel for the model. This is my first experiment with U-NET.

Whether I run the model from a DIGITS docker image or from the DIGITS digits-devserver I built from source, I get the same error:

ParseError 1:1 : Message type "caffe.NetParameter" has no field named "from".

Here are my NGC DIGITS docker images:

$ docker images

REPOSITORY TAG IMAGE ID CREATED SIZE nvidia/digits latest 32ba009273c8 7 days ago 2.8GB nvcr.io/nvidia/digits 18.02 4f7dc6bf79a6 3 weeks ago 5.45GB

FWIW I am using the /nvidia/digits image as the nvcr.io/nvidia/digits images runs for a few seconds then exits gracefully without any error messages generated in the terminal (I run it with the interactive (-it) flag). That issue will be the topic of another ticket.

Here are some details from the failed runs:

TRAINING DATA:

A. FEATURES Folder:

---> 30 single channel 572x572 grayscale PNG images. Each PNGis a single channel, 8-bit
grayscale image

---> Images are named from 01.png to 30.png

---> Each image file's properties were verfied via file and identify on the command line, e.g.:

  $ file 01.png
  01.png: PNG image data, 572 x 572, 8-bit grayscale, non-interlaced

B. LABELS Folder:

---> Same PNG image specifications and naming convention as above

---> Each image is a binary mask with white (255) labeling ground truth segmentation and black (0) labeling background

DATASET & SEGMENTATION MODEL CREATION:

C. NEW DATASET:

D. NEW SEGMENTATION MODEL:

E. ERROR:

All comments/critiques are appreciated.

[MiralPatel3011]

i have same issue. how you resolve it. Thank you