mirror of https://github.com/open-mmlab/mmocr.git
20 lines
732 B
Python
20 lines
732 B
Python
# The transcription of NAF dataset is annotated from Tessaract OCR, which is
|
|
# not accurate. The test/valid set ones were hand corrected, but the train set
|
|
# was only hand corrected a little. They aren't very good results. Better
|
|
# not to use them for recognition and text spotting.
|
|
|
|
_base_ = ['textdet.py']
|
|
data_root = 'data/naf'
|
|
|
|
data_converter = dict(
|
|
type='TextRecogCropConverter',
|
|
parser=dict(
|
|
type='NAFAnnParser', data_root=data_root, ignore=['¿', '§'],
|
|
det=False),
|
|
delete=['temp_images', 'naf_anno', 'data_split.json', 'annotations'])
|
|
|
|
config_generator = dict(
|
|
type='TextRecogConfigGenerator',
|
|
data_root=data_root,
|
|
val_anns=[dict(ann_file='textrecog_val.json', dataset_postfix='')])
|