Add tests #13
Add tests #13
Conversation
| labels_x = np.asarray([1, 1, 1, 0, 0, 2, 2, 4, 4, 4, 0], dtype=np.uint8) | ||
| labels_y = np.asarray([1, 1, 2, 0, 0, 3, 3, 4, 4, 4, 0], dtype=np.uint8) | ||
|
|
||
| reference_y_matched = np.asarray([1, 1, 5, 0, 0, 2, 2, 4, 4, 4, 0], dtype=np.uint8) |
There was a problem hiding this comment.
Hi Johannes @jo-mueller
could you please take a look at that test? I think this should fail. I think, the "5" should be a "1" in the reference. What do you think?
There was a problem hiding this comment.
Yes, reference_y_matched should essentially be a copy of labels_x here.
The "5" is actually correct and this is a good test because it covers a common situation: A label in slice_x may have been classified as two different labels in slice_y. match_labels is only allowed to merge labels in subsequent slices (x & y) if they have a large intersection, but it is not allowed to merge two labels in the same plane (i.e., x & x).
Thus, the labels with the largest overlap with 1 in labels_y will be matched (although they are already "1"), but the label that also overlaps the "1" in labels_x (i.e., label "2" in labels_y) will be labelled as a new label in the matched output since the label "2" is already occupied in labels_x.
This could be improved in future versions, but will require a much more elaborate approach than just measuring IoU.
There was a problem hiding this comment.
Yes, no I got it! Thanks :-)
|
I added another test script for /label/_label_overlap_matrix.py - could you also have a look at it? @haesleinhuepf |
|
Great! That's a good additional test. |
Description
This is just adding some more tests. I would like to make sure the functions do what they are supposed to do
Type of change
References
Tests
Final checks