🚧 Process lines

ocrd_calamari/cli.py

@@ -0,0 +1,10 @@
+import click
+
+from ocrd.decorators import ocrd_cli_options, ocrd_cli_wrap_processor
+from ocrd_calamari.ocr import CalamariOcr
+
+
+@click.command()
+@ocrd_cli_options
+def ocrd_calamari_ocr(*args, **kwargs):
+    return ocrd_cli_wrap_processor(CalamariOcr, *args, **kwargs)

ocrd_calamari/config.py

@@ -0,0 +1,4 @@
+import json
+from pkg_resources import resource_string
+
+OCRD_TOOL = json.loads(resource_string(__name__, 'ocrd-tool.json').decode('utf8'))

ocrd_calamari/ocr.py

@@ -1,18 +1,46 @@
 from __future__ import absolute_import
-from calamari_ocr.scripts.predict import run
 
-log = getLogger('processor.KrakenOcr')
+from glob import glob
 
-class KrakenOcr(Processor):
+import numpy as np
+from calamari_ocr.ocr import MultiPredictor
+from calamari_ocr.ocr.voting import voter_from_proto
+from calamari_ocr.proto import VoterParams
+from ocrd import Processor
+from ocrd.logging import getLogger
+from ocrd.model import ocrd_page
+from ocrd.utils import polygon_from_points
+
+from ocrd_calamari.config import OCRD_TOOL
+
+log = getLogger('processor.CalamariOcr')
+
+# TODO: Should this be "recognize", not "ocr" akin ocrd_tesserocr?
+
+
+class CalamariOcr(Processor):
 
     def __init__(self, *args, **kwargs):
         kwargs['ocrd_tool'] = OCRD_TOOL['tools']['ocrd-calamari-ocr']
-        super(KrakenOcr, self).__init__(*args, **kwargs)
+        super(CalamariOcr, self).__init__(*args, **kwargs)
+
+
+    def _init_calamari(self):
+        checkpoints = glob('/home/mike/devel/experiments/train-calamari-gt4histocr/models/*.ckpt.json')  # XXX
+        self.predictor = MultiPredictor(checkpoints=checkpoints)
+
+        voter_params = VoterParams()
+        voter_params.type = VoterParams.Type.Value('confidence_voter_default_ctc'.upper())
+        self.voter = voter_from_proto(voter_params)
+
 
     def process(self):
         """
-        Performs the binarization.
+        Performs the recognition.
         """
+
+        self._init_calamari()
+
         for (n, input_file) in enumerate(self.input_files):
             log.info("INPUT FILE %i / %s", n, input_file)
             pcgts = ocrd_page.from_file(self.workspace.download_file(input_file))
@@ -20,20 +48,22 @@ class KrakenOcr(Processor):
             log.info("pcgts %s", pcgts)
             for region in pcgts.get_Page().get_TextRegion():
                 textlines = region.get_TextLine()
-                log.info("About to binarize %i lines of region '%s'", len(textlines), region.id)
+                log.info("About to recognize %i lines of region '%s'", len(textlines), region.id)
                 for (line_no, line) in enumerate(textlines):
-                    log.debug("Binarizing line '%s' in region '%s'", line_no, region.id)
-                    image = self.workspace.resolve_image_as_pil(image_url, polygon_from_points(line.get_Coords().points))
-                    print(dir(kraken.binarization))
-                    bin_image = kraken.binarization.nlbin(image)
-                    bin_image_bytes = io.BytesIO()
-                    bin_image.save(bin_image_bytes, format='PNG')
-                    ID = concat_padded(self.output_file_grp, n)
-                    self.add_output_file(
-                        ID=ID,
-                        file_grp=self.output_file_grp,
-                        basename="%s.bin.png" % ID,
-                        mimetype='image/png',
-                        content=bin_image_bytes.getvalue()
-                    )
+                    log.debug("Recognizing line '%s' in region '%s'", line_no, region.id)
+                    image = self.workspace.resolve_image_as_pil(image_url,
+                                                                polygon_from_points(line.get_Coords().points))
+                    image_np = np.array(image, dtype=np.uint8)  # XXX better way?
 
+                    raw_results = list(self.predictor.predict_raw([image_np], progress_bar=False))[0]
+
+                    for i, p in enumerate(raw_results):
+                        p.prediction.id = "fold_{}".format(i)
+
+                    prediction = self.voter.vote_prediction_result(raw_results)
+                    prediction.id = "voted"
+
+                    print('***', prediction.sentence)
+                    print(prediction.avg_char_probability)
+                    for raw_result in raw_results:
+                        print(raw_result.sentence)

ocrd_calamari/ocrd-tool.json

@@ -10,13 +10,19 @@
       "steps": [
         "recognition/text-recognition"
       ],
-      "description": "Recognize lines with kraken",
+      "description": "Recognize lines with Calamari",
+      "input_file_grp": [
+        "OCR-D-SEG-LINE"
+      ],
+      "output_file_grp": [
+        "OCR-D-OCR-CALAMARI"
+      ],
       "parameters": {
-        "checkpoint": {"type": "string", "format": "file", "cacheable": true},
+        "XXX checkpoint": {"type": "string", "format": "file", "cacheable": true},
         "processes": {"type": "number", "default": 1},
         "batch_size": {"type": "number", "default": 1},
-        "voter": {"type": "string", "default": "confidence_voter_default_ctc"},
-        "extended_prediction_data_format": {"type": "string", "default": "json"},
+        "XXX voter": {"type": "string", "default": "confidence_voter_default_ctc"},
+        "XXXX extended_prediction_data_format": {"type": "string", "default": "json"},
         "XXX output_dir": "TODO",
         "XXX extended_prediction_data": "TODO"
       }