.pre-commit-config.yaml

@@ -7,7 +7,7 @@ repos:
     rev: 22.3.0
     hooks:
     - id: black
-  - repo: https://gitlab.com/pycqa/flake8
+  - repo: https://github.com/pycqa/flake8
     rev: 3.9.2
     hooks:
       - id: flake8

kaldi_data_generator/arguments.py

@@ -31,6 +31,7 @@ class TranscriptionType(Enum):
     paragraph: str = "paragraph"
     act: str = "act"
     page: str = "page"
+    text: str = "text"
 
 
 @dataclass
@@ -136,6 +137,7 @@ class FilterArgs:
             Use `--accepted_worker_version_ids manual` to get only manual transcriptions
         style: Filter line images by style class. 'other' corresponds to line elements that have neither
             handwritten or typewritten class : {[s.name for s in Style]}
+        accepted_metadatas: Key-value dictionary where each entry is a mandatory Arkindex metadata name/value. Filter lines by metadata.
     """
 
     transcription_type: TranscriptionType = TranscriptionType.text_line
@@ -144,3 +146,4 @@ class FilterArgs:
     accepted_worker_version_ids: List[str] = field(default_factory=list)
     skip_vertical_lines: bool = False
     style: Style = None
+    accepted_metadatas: dict = field(default_factory=dict)

kaldi_data_generator/main.py

@@ -81,6 +81,8 @@ class HTRDataGenerator:
         self.should_filter_by_worker = bool(self.accepted_worker_version_ids)
         self.style = filter.style
         self.should_filter_by_style = bool(self.style)
+        self.accepted_metadatas = filter.accepted_metadatas
+        self.should_filter_by_metadatas = bool(self.accepted_metadatas)
         self.transcription_type = filter.transcription_type.value
         self.skip_vertical_lines = filter.skip_vertical_lines
         self.skipped_pages_count = 0
@@ -150,11 +152,25 @@ class HTRDataGenerator:
         img = read_img(cached_img_path, self.grayscale)
         return img
 
+    def metadata_filtering(self, elt):
+        metadatas = {
+            metadata["name"]: metadata["value"] for metadata in elt["metadata"]
+        }
+        for meta in self.accepted_metadatas:
+            if not (
+                meta in metadatas and metadatas[meta] == self.accepted_metadatas[meta]
+            ):
+                return False
+        return True
+
     def get_accepted_zones(self, page_id: str):
         try:
             accepted_zones = []
             for elt in self.api_client.cached_paginate(
-                "ListElementChildren", id=page_id, with_classes=True
+                "ListElementChildren",
+                id=page_id,
+                with_classes=self.should_filter_by_class,
+                with_metadata=self.should_filter_by_metadatas,
             ):
                 elem_classes = [c for c in elt["classes"] if c["state"] != "rejected"]
 
@@ -193,12 +209,17 @@ class HTRDataGenerator:
                         raise ValueError(
                             f"Multiple style classes on the same element! {elt['id']} - {elem_classes}"
                         )
+                        should_accept = found_class == self.style
+
+                if not should_accept:
+                    continue
+
+                if self.should_filter_by_metadatas:
+                    if self.metadata_filtering(elt):
 
-                    if found_class == self.style:
                         accepted_zones.append(elt["zone"]["id"])
                 else:
                     accepted_zones.append(elt["zone"]["id"])
-
             logger.info(
                 "Number of accepted zone for page {} : {}".format(
                     page_id, len(accepted_zones)
@@ -269,9 +290,11 @@ class HTRDataGenerator:
                     and res["worker_version_id"] not in self.accepted_worker_version_ids
                 ):
                     continue
-                if (self.should_filter_by_class or self.should_filter_by_style) and (
-                    res["element"]["zone"]["id"] not in accepted_zones
-                ):
+                if (
+                    self.should_filter_by_class
+                    or self.should_filter_by_style
+                    or self.should_filter_by_metadatas
+                ) and (res["element"]["zone"]["id"] not in accepted_zones):
                     continue
                 if res["element"]["type"] != self.transcription_type:
                     continue
@@ -280,7 +303,7 @@ class HTRDataGenerator:
                 if not text or not text.strip():
                     continue
 
-                if "\n" in text.strip():
+                if "\n" in text.strip() and not self.transcription_type == "text":
                     elem_id = res["element"]["id"]
                     raise ValueError(
                         f"Newlines are not allowed in line transcriptions - {page_id} - {elem_id} - {text}"
@@ -396,7 +419,11 @@ class HTRDataGenerator:
             )
 
     def extract_lines(self, page_id: str, image_data: dict):
-        if self.should_filter_by_class or self.should_filter_by_style:
+        if (
+            self.should_filter_by_class
+            or self.should_filter_by_style
+            or self.should_filter_by_metadatas
+        ):
             accepted_zones = self.get_accepted_zones(page_id)
         else:
             accepted_zones = []

tests/conftest.py

@@ -41,6 +41,23 @@ def fake_image():
     return extractor.read_img(img_path)
 
 
+@pytest.fixture
+def fake_run_filter_metadata():
+    api_client = MockApiClient()
+    # ignore caching for this test
+    api_client.cached_paginate = api_client.paginate
+    with open(FIXTURES / "Maurdor/ListElementChildren/fake_page.json") as f:
+        pages_json = json.load(f)
+        api_client.add_response(
+            "ListElementChildren",
+            response=pages_json,
+            id="fake_page",
+            with_classes=False,
+            with_metadata=True,
+        )
+    return api_client
+
+
 @pytest.fixture
 def fake_run_volume_api_client(fake_volume_id):
     api_client = MockApiClient()

tests/data/Maurdor/ListElementChildren/fake_page.json

+[
+    {
+        "id": "0b719a5a-40c7-47ec-96c8-6c3064df5485", "type": "text", "name": "29", "corpus": {"id": "809222f2-b4a4-444c-a1a8-37667ccbff6b", "name": "Maurdor", "public": false}, "thumbnail_url": null, "zone": {"id": "0b719a5a-40c7-47ec-96c8-6c3064df5485", "polygon": [[190, 1187], [190, 1242], [389, 1242], [389, 1187], [190, 1187]], "image": {"id": "65955cb6-7aeb-4f7c-8531-5d797c135e41", "path": "public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "width": 1700, "height": 2339, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "s3_url": null, "status": "checked", "server": {"display_name": "https://europe-gamma.iiif.teklia.com/iiif/2", "url": "https://europe-gamma.iiif.teklia.com/iiif/2", "max_width": null, "max_height": null}}, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png/190,1187,199,55/full/0/default.jpg"}, "rotation_angle": 0, "mirrored": false, "created": "2022-09-05T22:15:11.027308Z", "classes": [], "metadata": [{"id": "8b5100cf-98c6-4d0f-8bbe-6df07cbdd02e", "type": "text", "name": "Language", "value": "arabic", "dates": []}, {"id": "bf7de461-1ca2-4160-b1c7-be63e28bd06e", "type": "text", "name": "Script", "value": "typed", "dates": []}], "has_children": null, "worker_version_id": "329f5b6e-78c9-4240-a2cf-78a746c6f897", "confidence": null
+    },
+    {"id": "fbf0d8e5-4729-4bd1-988a-49e178f7d0e6", "type": "text", "name": "27", "corpus": {"id": "809222f2-b4a4-444c-a1a8-37667ccbff6b", "name": "Maurdor", "public": false}, "thumbnail_url": null, "zone": {"id": "fbf0d8e5-4729-4bd1-988a-49e178f7d0e6", "polygon": [[676, 2124], [676, 2195], [1070, 2195], [1070, 2124], [676, 2124]], "image": {"id": "65955cb6-7aeb-4f7c-8531-5d797c135e41", "path": "public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "width": 1700, "height": 2339, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "s3_url": null, "status": "checked", "server": {"display_name": "https://europe-gamma.iiif.teklia.com/iiif/2", "url": "https://europe-gamma.iiif.teklia.com/iiif/2", "max_width": null, "max_height": null}}, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png/676,2124,394,71/full/0/default.jpg"}, "rotation_angle": 0, "mirrored": false, "created": "2022-09-05T22:15:11.027161Z", "classes": [], "metadata": [{"id": "75f36675-92a1-43df-90d0-bccd97b43594", "type": "text", "name": "Function", "value": "reference", "dates": []}, {"id": "e90372ce-db83-47ca-9362-b2e5def497e8", "type": "text", "name": "Language", "value": "english", "dates": []}, {"id": "bd1ad52f-dbb4-4dc8-a865-d6c42abd690e", "type": "text", "name": "Script", "value": "typed", "dates": []}], "has_children": null, "worker_version_id": "329f5b6e-78c9-4240-a2cf-78a746c6f897", "confidence": null},
+    {"id": "18725942-24f5-4f81-a16f-62c1323c1041", "type": "text", "name": "28", "corpus": {"id": "809222f2-b4a4-444c-a1a8-37667ccbff6b", "name": "Maurdor", "public": false}, "thumbnail_url": null, "zone": {"id": "18725942-24f5-4f81-a16f-62c1323c1041", "polygon": [[1154, 1244], [1154, 1327], [1456, 1327], [1456, 1244], [1154, 1244]], "image": {"id": "65955cb6-7aeb-4f7c-8531-5d797c135e41", "path": "public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "width": 1700, "height": 2339, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "s3_url": null, "status": "checked", "server": {"display_name": "https://europe-gamma.iiif.teklia.com/iiif/2", "url": "https://europe-gamma.iiif.teklia.com/iiif/2", "max_width": null, "max_height": null}}, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png/1154,1244,302,83/full/0/default.jpg"}, "rotation_angle": 0, "mirrored": false, "created": "2022-09-05T22:15:11.027234Z", "classes": [], "metadata": [{"id": "1b78325e-24fa-452f-b11c-5db2ac52df59", "type": "text", "name": "Language", "value": "arabic", "dates": []}, {"id": "85ce21e0-0521-4e59-bfd0-a2060888be56", "type": "text", "name": "Script", "value": "typed", "dates": []}], "has_children": null, "worker_version_id": "329f5b6e-78c9-4240-a2cf-78a746c6f897", "confidence": null},
+    {"id": "3ebe7e48-fe2f-4533-92df-9895db05c3f5", "type": "text", "name": "23", "corpus": {"id": "809222f2-b4a4-444c-a1a8-37667ccbff6b", "name": "Maurdor", "public": false}, "thumbnail_url": null, "zone": {"id": "3ebe7e48-fe2f-4533-92df-9895db05c3f5", "polygon": [[1403, 1763], [1403, 1830], [1470, 1830], [1470, 1763], [1403, 1763]], "image": {"id": "65955cb6-7aeb-4f7c-8531-5d797c135e41", "path": "public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "width": 1700, "height": 2339, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png", "s3_url": null, "status": "checked", "server": {"display_name": "https://europe-gamma.iiif.teklia.com/iiif/2", "url": "https://europe-gamma.iiif.teklia.com/iiif/2", "max_width": null, "max_height": null}}, "url": "https://europe-gamma.iiif.teklia.com/iiif/2/public%2Fmaurdor%2Fdev_png%2FYXLOKV-00.png/1403,1763,67,67/full/0/default.jpg"}, "rotation_angle": 0, "mirrored": false, "created": "2022-09-05T22:15:11.026869Z", "classes": [], "metadata": [{"id": "42034ac4-1ed3-4893-a275-296484b417c5", "type": "text", "name": "Language", "value": "arabic", "dates": []}, {"id": "d20a52bb-28e7-407e-998e-b51f10af330a", "type": "text", "name": "Script", "value": "typed", "dates": []}], "has_children": null, "worker_version_id": "329f5b6e-78c9-4240-a2cf-78a746c6f897", "confidence": null}
+
+
+]
\ No newline at end of file

tests/test_main.py

@@ -58,6 +58,23 @@ def test_run_volumes_with_worker_version(
     assert api_client.responses == []
 
 
+def test_get_accepted_zones_filter_metadata(tmpdir, fake_run_filter_metadata):
+    api_client = fake_run_filter_metadata
+    out_dir_base = tmpdir
+    htr_data_gen = HTRDataGenerator(
+        format="kaldi",
+        filter=FilterArgs(accepted_metadatas={"Language": "arabic"}),
+        out_dir_base=out_dir_base,
+    )
+    htr_data_gen.api_client = api_client
+
+    assert htr_data_gen.get_accepted_zones("fake_page") == [
+        "0b719a5a-40c7-47ec-96c8-6c3064df5485",
+        "18725942-24f5-4f81-a16f-62c1323c1041",
+        "3ebe7e48-fe2f-4533-92df-9895db05c3f5",
+    ]
+
+
 def test_create_partitions(fake_expected_partitions, tmpdir):
     out_dir_base = Path(tmpdir)
     splitter = KaldiPartitionSplitter(out_dir_base=out_dir_base)