search

Deci-AI / super-gradients

Easily train or fine-tune SOTA computer vision models with one open source training library. The home of Yolo-NAS.
https://www.supergradients.com
Apache License 2.0
4.54k stars 496 forks source link

ReverseImageChannels transform's get_equivalent_preprocessing return unwanted set object #1122

Closed yitelee closed 1 year ago

yitelee commented 1 year ago

/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 commented 1 year ago

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.