qurator-spk/eynollah
1
0
Fork 0
mirror of https://github.com/qurator-spk/eynollah.git synced 2026-04-30 19:22:03 +02:00
Code Issues Projects Releases Wiki Activity

.

Browse source
This commit is contained in:
kba 2026-04-28 14:51:23 +02:00
parent 5c6e075975
commit 511222704e
4 changed files with 53 additions and 43 deletions
Download patch file Download diff file Expand all files Collapse all files

2
src/eynollah/eynollah_ocr.py
View file

@ -72,7 +72,7 @@ class Eynollah_ocr:
self.model_zoo.load_model('ocr', '')
self.model_zoo.load_model('num_to_char')
self.model_zoo.load_model('characters')
self.end_character = len(self.model_zoo.get('characters', list)) + 2
self.end_character = len(self.model_zoo.get('characters')) + 2
@property
def device(self):

9
src/eynollah/model_zoo/model_zoo.py
View file

@ -197,9 +197,12 @@ class EynollahModelZoo:
return model
def get(self, model_category: str) -> Predictor:
if model_category not in self._loaded:
raise ValueError(f'Model "{model_category}" not previously loaded with "load_model(..)"')
return self._loaded[model_category]
# if model_category not in self._loaded:
# raise ValueError(f'Model "{model_category}" not previously loaded with "load_model(..)"')
if model_category in self._loaded:
return self._loaded[model_category]
else:
return self.load_model(model_category)
def _load_ocr_model(self, variant: str) -> AnyModel:
"""

81
src/eynollah/ocrd-tool.json
View file

@ -162,43 +162,50 @@
"version_range": "< v0.7.0"
}
]
}
},
"ocrd-eynollah-recognize": {
"executable": "ocrd-eynollah-recognize",
"categories": ["Text recognition and optimization"],
"steps": ["recognition/text-recognition"],
"input_file_grp_cardinality": 1,
"output_file_grp_cardinality": 1,
"parameters": {
"models": {
"type": "string",
"format": "uri",
"content-type": "text/directory",
"cacheable": true,
"description": "Directory containing the eynollah_models directory",
"required": true
},
"do_not_mask_with_textline_contour": {
"type": "boolean",
"description": "if this parameter set to true, cropped textline images will not be masked with textline contour.",
"default": false
},
"tr_ocr": {
"type": "boolean",
"description": "Whether to use (much more resource-intensive) transformer model",
"default": false
}
},
"resources": [
{
"url": "https://zenodo.org/records/17580627/files/models_ocr_v0_6_0.tar.gz?download=1",
"name": "models_ocr_v0_6_0",
"type": "archive",
"size": 6119874002,
"description": "Models for OCR",
"version_range": ">= v0.6.0"
}
]
"ocrd-eynollah-recognize": {
"executable": "ocrd-eynollah-recognize",
"categories": ["Text recognition and optimization"],
"steps": ["recognition/text-recognition"],
"input_file_grp_cardinality": 1,
"output_file_grp_cardinality": 1,
"description": "Recognize text with eynollah (CNN/RNN or Transformer)",
"parameters": {
"models": {
"type": "string",
"format": "uri",
"content-type": "text/directory",
"cacheable": true,
"description": "Directory containing the eynollah_models directory",
"required": true
},
"do_not_mask_with_textline_contour": {
"type": "boolean",
"description": "if this parameter set to true, cropped textline images will not be masked with textline contour.",
"default": false
},
"tr_ocr": {
"type": "boolean",
"description": "Whether to use (much more resource-intensive) transformer model",
"default": false
},
"batch_size": {
"type": "number",
"format": "integer",
"description": "Batch size, leave as 0 for builtin default (2 for CNN/RNN, 2 for TrOCR)",
"default": 0
}
},
"resources": [
{
"url": "https://zenodo.org/records/17580627/files/models_ocr_v0_6_0.tar.gz?download=1",
"name": "models_ocr_v0_6_0",
"type": "archive",
"size": 6119874002,
"description": "Models for OCR",
"version_range": ">= v0.6.0"
}
]
}
}
}

4
src/eynollah/ocrd_cli_recognize.py
View file

@ -34,8 +34,8 @@ class EynollahRecognizeProcessor(Processor):
model_zoo=model_zoo,
tr_ocr=self.parameter['tr_ocr'],
do_not_mask_with_textline_contour=self.parameter['do_not_mask_with_textline_contour'],
batch_size=self.parameter['batch_size'],
min_conf_value_of_textline_text=self.parameter['min_conf_value_of_textline_text'])
batch_size=self.parameter['batch_size'] if self.parameter['batch_size'] >= 0 else 2 if self.parameter['tr_ocr'] else 8,
min_conf_value_of_textline_text=0)
# FIXME: This is just a proof-of-concept, very inefficient and non-conformant
# TODO: OCR writing should use PAGE API once result dataclass mechanism is settled,