Add tests for Cityscapes Dataset #1079
Conversation
|
Can you add a test for having multiple target types, e.g semantic and polygon? |
Codecov Report
@@ Coverage Diff @@
## master #1079 +/- ##
==========================================
+ Coverage 63.92% 64.52% +0.59%
==========================================
Files 68 68
Lines 5406 5406
Branches 829 829
==========================================
+ Hits 3456 3488 +32
+ Misses 1707 1666 -41
- Partials 243 252 +9
Continue to review full report at Codecov.
|
|
@TheCodez yep can do +1 |
|
@fmassa I'll do it in this PR I think +1 |
| def cityscapes_root(): | ||
|
|
||
| def _make_image(file): | ||
| PIL.Image.fromarray(np.zeros((1024, 2048, 3), dtype=np.uint8)).save(file) |
There was a problem hiding this comment.
I don't know how big the impact is but maybe we could use a smaller image size here instead of the original size to make the tests run faster?
There was a problem hiding this comment.
We are not performing any significant operations on the inputs, so this should not be a problem speed-wise.
But good point!
semantic,instance,polygon,color) and modes (gtFine,gtCoarse).generic_segmentation_dataset_test- checks the outputs (image and mask) are bothPIL.Imagetarget_dir_zipfor coarse mode (discovered with tests ;))Tried to base this on the style of the other dataset tests - it's a bit more involved with Cityscapes mocking, but let me know if you have any further ideas/refinements for testing and I can incorporate.