Segmentation model detections are all regular shapes

[grantporterdg]

This is happening with all of the current tests with segmentation models.

  Scenario: Local detect using segmentation models using default settings (GPU)
  User launches OSN using a segmentation model with a local image using default settings

    Given I am at a new command line
    When I issue the command
            """
            ../bin/OpenSpaceNet --model ../data/SegwayNet.gbdxm
            --type POLYGON --format shp --output poly_shp/detection.shp
            --image ../data/dickerson_R1C1.tif
            """
    Then The return code is "0"
    And we see only the caffe warning in stderr
    And the number of files created should be "6"
    And a shapefile named "poly_shp/detection.shp" should be created
    And GDAL says there are about "14" features in the file "poly_shp/detection.shp"
    And Displayed feature count agrees with GDAL feature count

output is regular shapes and lots of nulls in the data