Hi all,
I'm trying to fine-tuning DONUT model over my dataset for the task Key Information Extraction. My parsing dictionary is similar to SROIE but for all the sample there are at least 1 key with has a missing value.
Is it better to train with a ground truth like:
{ 'key1': 'nan', 'key2': 'value2', 'key3': 'value3', 'key4': 'nan'}
or just
{'key2': 'value2', 'key3': 'value3'}
Hi all, I'm trying to fine-tuning DONUT model over my dataset for the task Key Information Extraction. My parsing dictionary is similar to SROIE but for all the sample there are at least 1 key with has a missing value. Is it better to train with a ground truth like: { 'key1': 'nan', 'key2': 'value2', 'key3': 'value3', 'key4': 'nan'} or just {'key2': 'value2', 'key3': 'value3'}
Thanks for the answers!