To load the data, download the Omniglot dataset [link] and
extract both the images_background.zip and images_evaluation.zip into the same folder. The dataset is
originally designed for one-shot learning, the train/valid splitting can not be used in our Multi-Task Learning
experiments. To generate the dataset splitting, we should first change the dataTool.MAIN_DIR to
the path of the dataset and run dataTool.convert_paths() to randomly split the images of
each task (language) into train and test sets.
Change the TAAN.MAIN_DIR to the path of the dataset and run:
python Evaluation_TAAN.py --regularize cosine --gpu 0