fixed Mask2Former image processor segmentation maps handling #33364
Conversation
5c73b13 to
7aad102
Compare
| image_mean=0.5, | ||
| image_std=0.5, |
There was a problem hiding this comment.
Please, revert to the original state :)
There was a problem hiding this comment.
When a revert is applied both values will become 3-element lists and will not be compatible with tests run on images with a different number of channels.
There was a problem hiding this comment.
If we want to be super precise I can write '[0.5] * num_channels'.
There was a problem hiding this comment.
Yes, that way will be better, just to be sure normalization work for this case 👍
| with pytest.raises(ValueError, match="Unable to infer channel dimension format"): | ||
| common(num_channels=5, numpify=True, do_resize=False) | ||
|
|
||
| with pytest.raises(TypeError, match=r"Cannot handle this data type: .*"): | ||
| common(num_channels=5, numpify=True, input_data_format=ChannelDimension.LAST) |
There was a problem hiding this comment.
Sorry, missed this for the first time, we use unittest for testing, can you please update to with self.assertRaises(ValueError) or assertRaisesRegex and remove import pytest on the top of the module.
| self.image_processor_tester.num_channels = num_channels | ||
| self.image_processor_tester.do_resize = do_resize | ||
| self.image_processor_tester.image_mean = [0.5] * num_channels | ||
| self.image_processor_tester.image_std = [0.5] * num_channels |
There was a problem hiding this comment.
not sure it's a good way to modify class properties in place, that may lead to some hard-to-debug issues later. You can recreate an instance here with an updated image processor dict.
| common(num_channels=1, numpify=True) | ||
| common(num_channels=2, numpify=True, input_data_format=ChannelDimension.LAST) | ||
| common(num_channels=5, numpify=True, input_data_format=ChannelDimension.LAST, do_resize=False) |
There was a problem hiding this comment.
Can you please make a call with explicit input_data_format=ChannelDimension.FIRST for better readability of the test?
There was a problem hiding this comment.
Thanks for fixing! With @qubvel's suggestions all looks good to me -- let's get his final approval and then we can merge :)
…face#33364) * fixed mask2former image processor segmentation maps handling * introduced review suggestions * introduced review suggestions
Fixes segmentation maps padding size calculation when handling multichannel images using previously inferred
input_data_format.Fixes #33295
Before submitting
Pull Request section?
to it if that's the case -> Mask2FormerImageProcessor - fails to process multichannel image #33295.
Who can review?
Anyone