Add random_sharpness processing layer #20697
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #20697 +/- ##
=======================================
Coverage 81.95% 81.95%
=======================================
Files 546 547 +1
Lines 50966 51036 +70
Branches 7884 7893 +9
=======================================
+ Hits 41768 41826 +58
- Misses 7278 7285 +7
- Partials 1920 1925 +5
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
| def transform_images(self, images, transformation=None, training=True): | ||
| if training: | ||
| if self.data_format == "channels_first": | ||
| images = self.backend.numpy.swapaxes(images, -3, -1) |
There was a problem hiding this comment.
{{function_node __wrapped__DepthwiseConv2dNative_device_/job:localhost/replica:0/task:0/device:CPU:0}} Depthwise convolution on CPU is only supported for NHWC format [Op:DepthwiseConv2dNative] name:
To resolve this error, I adjusted the image axis based on the data_format to ensure compatibility with the depthwise_conv operation. This adjustment was necessary because depthwise convolution on CPU is only supported in the NHWC format.
There was a problem hiding this comment.
Your approach of transposing to channels_last and then transposing back again is good.
| def transform_images(self, images, transformation=None, training=True): | ||
| if training: | ||
| if self.data_format == "channels_first": | ||
| images = self.backend.numpy.swapaxes(images, -3, -1) |
There was a problem hiding this comment.
Your approach of transposing to channels_last and then transposing back again is good.
google-ml-butler
bot
added
kokoro:force-run
ready to pull
google-ml-butler
bot
removed
ready to pull
I implemented the RandomSharpness image processing feature before working on rand_augment, as it includes related image processing functionalities.
Here is my gist