chunked/nltk_chunking/tasks_image_classification/chunk_8.txt

from transformers import DefaultDataCollator
data_collator = DefaultDataCollator()

To avoid overfitting and to make the model more robust, add some data augmentation to the training part of the dataset.