ReverseImageChannels transform's get_equivalent_preprocessing return unwanted set object

[yitelee]

/usr/local/lib/python3.10/dist-packages/super_gradients/training/processing/processing.py in preprocess_image(self, image)
72 processed_image, metadata_lst = image.copy(), []
73 for processing in self.processings:
---> 74 processed_image, metadata = processing.preprocess_image(image=processed_image)
75 metadata_lst.append(metadata)
76 return processed_image, ComposeProcessingMetadata(metadata_lst=metadata_lst)

AttributeError: 'set' object has no attribute 'preprocess_image'

print(best_model._image_processor.processings)

[<super_gradients.training.processing.processing.ReverseImageChannels at 0x7fe14c69cd60>,
<super_gradients.training.processing.processing.DetectionLongestMaxSizeRescale at 0x7fe14c69fca0>,
{'ReverseImageChannels'},
<super_gradients.training.processing.processing.DetectionCenterPadding at 0x7fe14c69db40>,
<super_gradients.training.processing.processing.StandardizeImage at 0x7fe14c69dfc0>,
<super_gradients.training.processing.processing.ImagePermute at 0x7fe14c69ee30>]

https://github.com/Deci-AI/super-gradients/blame/b4eec926261e9e8f3032d8fb51977af33b85ec3c/src/super_gradients/training/transforms/transforms.py#L944

Tested OK if changed
return [{Processings.ReverseImageChannels}] --> return [Processings.ReverseImageChannels]

[aayushsingla]

Facing similar issue. Can reproduce with YoloNAS when using model.predict with image path or array of paths. model.predict works fine with a url though but throws same error with prediction.save.