What's Changed
* Further optimized Categorizer training to improve its accuracy and generalizibility.
* Significantly increased the data amount of augmentation when training a Categorizer. As a result, training a Categorizer requires less sorted behavior examples.
* Since the data amount of augmentation significantly increased, larger memory is required to train a Categorizer if loading all training examples into memory. Therefore, users can now export the augmented training examples to a folder and train a Categorizer without requiring a lot of memory but at slower speed. The exported training examples can also be directly used for future training without performing augmentation again.
* Implemented a sorting method that uses a '.csv' file that stores the frame-wise behavioral labels to sort the unsorted behavior examples generated by LabGym. As a result, users can use behavioral annotations from other tools to automatically sort behavior examples in LabGym.
* Added a 'draw lines' function to the 'Draw Markers' functional unit so that users can draw both lines and circles to mark a specific location in videos.
**Full Changelog**: <https://github.com/umyelab/LabGym/compare/v2.7.2...v2.8.0>