Skip to content

Latest commit

 

History

History
executable file
·
54 lines (33 loc) · 2.07 KB

CHANGELOG.md

File metadata and controls

executable file
·
54 lines (33 loc) · 2.07 KB
Raw

Changes

2021-09-06

  • Fixed the bug with ImageDataAugmentor.fit when custom targets have been specified.

  • Changed the signature of ImageDataAugmentor.preprocess_output to ImageDataAugmentor.preprocess_labels (the latter is more descriptive)

  • Changed the DataFrameIterator.class_mode options image_target and mask_target to better describing color_target and grayscale_target, respectively.

  • Finished the segmentation example ./examples/segmentation-with-flow_from_dataframe.ipynb

2020-12-21

  • Added ImageDataAugmentor.input_augment_mode that enables selecting augmentations to inputs

  • Logic for ImageDataAugmentor.input_augment_mode and ImageDataAugmentor.label_augment_mode parameters

  • Added an usage example for the aforementioned augment modes in README.md

  • Fixed some small bugs: unused kwargs will now throw an error, class_mode==None returnables fixed

2020-12-17

  • ImageDataAugmentor.label_augment_mode enables targeting augmentations to labels, e.g if ImageDataAugmentor.flow_from_dataframe.class_mode is set to image_target or mask_target

  • Iterator.seed parameter moved from iterators to ImageDataAugmentor.seed: deprecated usage will throw a warning.

  • ImageDataAugmentor.augment_seed has been removed and is now governed by ImageDataAugmentor.seed (see above)

  • ImageDataAugmentor.augment_mode parameter has been removed and the functionality replaced with ImageDataAugmentor.label_augment_mode (see above)

  • Iterator.gray added to color modes (a synonym for grayscale)

  • Iterator.show_batch has been replaced with Iterator.show_data

  • Added examples to codebase/examples

  • Iterator.dtype now casts the inputs and targets into the desired datatype right before the batch is being returned whereas earlier the datatype casting was done in the start of data generation. This will prevent errors caused by augmentations not being able to handle the casted datatypes.

  • Removed the direct support for imgaug since imgaug transformations can be called using augmentations.imgaug