yomitoku 0.7.2__py3-none-any.whl → 0.7.3__py3-none-any.whl

yomitoku/data/functions.py

@@ -132,7 +132,7 @@ def resize_shortest_edge(
     neww = max(int(new_w / 32) * 32, 32)
     newh = max(int(new_h / 32) * 32, 32)
 
-    img = cv2.resize(img, (neww, newh))
+    img = cv2.resize(img, (neww, newh), interpolation=cv2.INTER_AREA)
     return img
 
 
@@ -275,7 +275,7 @@ def resize_with_padding(img, target_size, background_color=(0, 0, 0)):
     new_w = int(w * min(scale_w, scale_h))
     new_h = int(h * min(scale_w, scale_h))
 
-    resized = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
+    resized = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_AREA)
 
     canvas = np.zeros((target_size[0], target_size[1], 3), dtype=np.uint8)
     canvas[:, :] = background_color

yomitoku/export/export_markdown.py

@@ -5,7 +5,7 @@ import cv2
 
 
 def escape_markdown_special_chars(text):
-    special_chars = r"([`*_{}[\]()#+.!|-])"
+    special_chars = r"([`*{}[\]()#+!~|-])"
     return re.sub(special_chars, r"\\\1", text)
 
 

yomitoku/models/parseq.py

@@ -81,6 +81,8 @@ class PARSeq(nn.Module, PyTorchModelHubMixin):
         named_apply(partial(init_weights, exclude=["encoder"]), self)
         nn.init.trunc_normal_(self.pos_queries, std=0.02)
 
+        self.export_onnx = False
+
     @property
     def _device(self) -> torch.device:
         return next(self.head.parameters(recurse=False)).device
@@ -175,7 +177,11 @@ class PARSeq(nn.Module, PyTorchModelHubMixin):
                     # greedy decode. add the next token index to the target input
                     tgt_in[:, j] = p_i.squeeze().argmax(-1)
                     # Efficient batch decoding: If all output words have at least one EOS token, end decoding.
-                    if testing and (tgt_in == self.tokenizer.eos_id).any(dim=-1).all():
+                    if (
+                        not self.export_onnx
+                        and testing
+                        and (tgt_in == self.tokenizer.eos_id).any(dim=-1).all()
+                    ):
                         break
 
             logits = torch.cat(logits, dim=1)

yomitoku/table_structure_recognizer.py

@@ -47,6 +47,7 @@ class TableStructureRecognizerSchema(BaseSchema):
     rows: List[TableLineSchema]
     cols: List[TableLineSchema]
     cells: List[TableCellSchema]
+    spans: List[TableLineSchema]
     order: int
 
 
@@ -242,7 +243,7 @@ class TableStructureRecognizer(BaseModule):
             category_elements
         )
 
-        cells, rows, cols = self.extract_cell_elements(category_elements)
+        cells, rows, cols, spans = self.extract_cell_elements(category_elements)
 
         table_x, table_y = data["offset"]
         table_x2 = table_x + data["size"][1]
@@ -255,6 +256,7 @@ class TableStructureRecognizer(BaseModule):
             "n_col": len(cols),
             "rows": rows,
             "cols": cols,
+            "spans": spans,
             "cells": cells,
             "order": 0,
         }
@@ -276,8 +278,9 @@ class TableStructureRecognizer(BaseModule):
 
         rows = sorted(elements["row"], key=lambda x: x["box"][1])
         cols = sorted(elements["col"], key=lambda x: x["box"][0])
+        spans = sorted(elements["span"], key=lambda x: x["box"][1])
 
-        return cells, rows, cols
+        return cells, rows, cols, spans
 
     def __call__(self, img, table_boxes, vis=None):
         img_tensors = self.preprocess(img, table_boxes)

yomitoku/text_recognizer.py

@@ -118,6 +118,7 @@ class TextRecognizer(BaseModule):
             "output": {0: "batch_size"},
         }
 
+        self.model.export_onnx = True
         torch.onnx.export(
             self.model,
             input,

{yomitoku-0.7.2.dist-info → yomitoku-0.7.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: yomitoku
-Version: 0.7.2
+Version: 0.7.3
 Summary: Yomitoku is an AI-powered document image analysis package designed specifically for the Japanese language.
 Author-email: Kotaro Kinoshita <kotaro.kinoshita@mlism.com>
 License: CC BY-NC-SA 4.0

{yomitoku-0.7.2.dist-info → yomitoku-0.7.3.dist-info}/RECORD

@@ -6,9 +6,9 @@ yomitoku/layout_analyzer.py,sha256=VhNf1ZQFoozj6WUGk5ll1p2p1jk5X3j-JPcDbTAoSl4,1
 yomitoku/layout_parser.py,sha256=V_mAkZxke1gwHfnxBFMTOJ8hnz2X_kfZu2lLiMd8cAs,7610
 yomitoku/ocr.py,sha256=JSTjkupcxHITQm6ERnzU7As0c3KWf8-oxc0AqNoWHXo,2272
 yomitoku/reading_order.py,sha256=OfhOS9ttPDoPSuHrIRKyOzG19GGeRufbuSKDqhsohh4,6404
-yomitoku/table_structure_recognizer.py,sha256=Eam9t7OjW4a-UWk_dl-ylbOcinN_Te_ovuri2naldL0,9482
+yomitoku/table_structure_recognizer.py,sha256=tHjex6deT_FjRK5ePz9bUXA_QIhgv_vYtK-ynm4ALxg,9625
 yomitoku/text_detector.py,sha256=XgqhtbNcJww2x3BrH8EFz45qC6kqPKCX9hsa-dzRoIA,4274
-yomitoku/text_recognizer.py,sha256=LVMjy-PaGlDQqfJrjKX_7vOQXDyFg6FaCeIQIyWUJX8,5833
+yomitoku/text_recognizer.py,sha256=t95sbxve-E9VOCaU9CFGZIlk_a4my9KfFfr9tXws9As,5871
 yomitoku/cli/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 yomitoku/cli/main.py,sha256=WvQO9V5HzxxvRAIsGkrDl9OGrmaKsAbDBrg4ApCSy_c,10527
 yomitoku/configs/__init__.py,sha256=e1Alss5QJLZSNfD6zLEG6xu5vDQDw-4Jayiqq8bq52s,571
@@ -19,15 +19,15 @@ yomitoku/configs/cfg_text_recognizer_parseq.py,sha256=hpFs3nKqh4XdU3BZMTultegtLE
 yomitoku/configs/cfg_text_recognizer_parseq_small.py,sha256=uCm_VC_G79IbZpOiK8fgYzAJ4b98H5pf328wyQomtfo,1259
 yomitoku/data/__init__.py,sha256=KAofFc9rk9ZdTKBjemu9RM8Vj9XnKbWC2MPZ2RWtOdE,82
 yomitoku/data/dataset.py,sha256=-I4f-FDtgsPnJ2MnXB7FtwihMW3koDaSI1OEoqKneIg,1014
-yomitoku/data/functions.py,sha256=eOyxo8S6EoAf1xGSPLWQFb9-t5Rg52NggD9MFIrOSpY,7506
+yomitoku/data/functions.py,sha256=7a_3xDKAQVdWfzQwFcdyJBojoyzUa3ePZOnG4pX1dpI,7532
 yomitoku/export/__init__.py,sha256=fkwOtqH0lh6eZQW5b4EMSjIH1FmWYLKKszahR-jQYSg,366
 yomitoku/export/export_csv.py,sha256=B234jlNeO4n5kQ_lwxxAZe_O2ipTbeDYlWU1zyyaVrw,3001
 yomitoku/export/export_html.py,sha256=pCLoxV10_SzRWmZlDnHuyfPFIuUGB3ZkqSdABVU7DTs,5038
 yomitoku/export/export_json.py,sha256=D6dD04gcPR5lmfHFVX-iGOYapsOVaJ_kH1Qhs6d2O0M,2035
-yomitoku/export/export_markdown.py,sha256=4mk_7P4kta6RguThnQUh7hesvZNLuXFnPlxhA951qVE,4171
+yomitoku/export/export_markdown.py,sha256=D1kX3X8odWa0pf4AFZ6gik5EKMKK7pgpQXaHHv6pWDI,4170
 yomitoku/models/__init__.py,sha256=Enxq9sjJWusZuxecTori8IQa8NEYKaiiptDluHX1avg,144
 yomitoku/models/dbnet_plus.py,sha256=jeWJZm0ihbxoJeAXBFK7uVIwoosx2IUNk7Ut5wRH0vA,7998
-yomitoku/models/parseq.py,sha256=-DQMQuON2jwtb4Ib2V0O19un9w-WG4rXS0SiscydrXU,8593
+yomitoku/models/parseq.py,sha256=psCPjP3eKjOFAUZJPQQhbD0nWEV5FeOZ0tTK27Rvvbw,8748
 yomitoku/models/rtdetr.py,sha256=oJsr8RHz3frslhLfXdVJve47lUsrmqLjfdTrZ41tlQ0,687
 yomitoku/models/layers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 yomitoku/models/layers/activate.py,sha256=S54GPssZBMloM2oFAXeDVMmBBZOWyjwU98Niq758txE,1244
@@ -48,7 +48,7 @@ yomitoku/utils/graph.py,sha256=LKNB8ZhSQwOZMfeAimPMF5UCVVr2ZaUWoGDkz8z-uGU,456
 yomitoku/utils/logger.py,sha256=uOmtQDr0A0JD7wyFshedL08BiNrQorHnpktRXba8bjU,424
 yomitoku/utils/misc.py,sha256=FbwPLeIYYBvNf9wQh2RoEonTM5BF7_IwaEqmRsYHKA8,2673
 yomitoku/utils/visualizer.py,sha256=DjDwHiAu1iFRKh96H3Egq4vuI2s_-9dLCDeykhKi8jo,5251
-yomitoku-0.7.2.dist-info/METADATA,sha256=4nyP_4zWDOxWgt7qfwRbgtlHL9xiv_FLzdw8YtfJNYU,8717
-yomitoku-0.7.2.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-yomitoku-0.7.2.dist-info/entry_points.txt,sha256=nFV3S11zgBNW0Qq_D0XQNg2R4lNXU_9XUFr6rdJoyF8,52
-yomitoku-0.7.2.dist-info/RECORD,,
+yomitoku-0.7.3.dist-info/METADATA,sha256=pHT4lQyl9cN4KbsOTooiJAaEEJqXhmAl9SVZKVaPkR0,8717
+yomitoku-0.7.3.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+yomitoku-0.7.3.dist-info/entry_points.txt,sha256=nFV3S11zgBNW0Qq_D0XQNg2R4lNXU_9XUFr6rdJoyF8,52
+yomitoku-0.7.3.dist-info/RECORD,,