Cleanup of PR #555 merge

kraken/lib/dataset/recognition.py

@@ -292,7 +292,7 @@ def __init__(self,
                  reorder: Union[bool, Literal['L', 'R']] = True,
                  im_transforms: Callable[[Any], torch.Tensor] = transforms.Compose([]),
                  augmentation: bool = False,
-                 legacy_polygons: bool=False) -> None:
+                 legacy_polygons: bool = False) -> None:
         """
         Creates a dataset for a polygonal (baseline) transcription model.
 

kraken/lib/dataset/segmentation.py

@@ -165,7 +165,6 @@ def __getitem__(self, idx):
                 im, target = self.transform(im, target)
                 return {'image': im, 'target': target}
             except Exception:
-                raise
                 self.failed_samples.add(idx)
                 idx = np.random.randint(0, len(self.imgs))
                 logger.debug(traceback.format_exc())

kraken/lib/pretrain/model.py

@@ -212,9 +212,10 @@ def __init__(self,
         if format_type == 'binary':
             legacy_train_status = train_set.legacy_polygons_status
             if val_set and val_set.legacy_polygons_status != legacy_train_status:
-                logger.warning(
-                    f'Train and validation set have different legacy polygon status: {legacy_train_status} and {val_set.legacy_polygons_status}.'
-                     'Train set status prevails.')
+                logger.warning('Train and validation set have different legacy '
+                               f'polygon status: {legacy_train_status} and '
+                               f'{val_set.legacy_polygons_status}. Train set '
+                               'status prevails.')
             if legacy_train_status == "mixed":
                 logger.warning('Mixed legacy polygon status in training dataset. Consider recompilation.')
                 legacy_train_status = False

kraken/lib/pretrain/util.py

@@ -139,7 +139,7 @@ def arrange(s, e, length, keep_length):
         for length in sorted(lengths, reverse=True):
             lens = np.fromiter(
                 (e - s if e - s >= length + mask_min_space else 0 for s, e in parts),
-                np.int,
+                int,
             )
             l_sum = np.sum(lens)
             if l_sum == 0:

kraken/lib/segmentation.py

@@ -25,6 +25,7 @@
 import torch
 import torch.nn.functional as F
 from PIL import Image, ImageDraw
+from PIL.Image import Resampling, Transform
 from scipy.ndimage import (binary_erosion, distance_transform_cdt,
                            gaussian_filter, maximum_filter, affine_transform)
 from scipy.signal import convolve2d
@@ -419,8 +420,8 @@ def _rotate(image: _T_pil_or_np,
     if isinstance(image, Image.Image):
         # PIL is much faster than scipy
         pdata = tform.params.flatten().tolist()[:6]
-        resample = {0: Image.NEAREST, 1: Image.BILINEAR, 2: Image.BICUBIC, 3: Image.BICUBIC}.get(order, Image.NEAREST)
-        return tform, image.transform(output_shape[::-1], Image.AFFINE, data=pdata, resample=resample, fillcolor=cval)
+        resample = {0: Resampling.NEAREST, 1: Resampling.BILINEAR, 2: Resampling.BICUBIC, 3: Resampling.BICUBIC}.get(order, Resampling.NEAREST)
+        return tform, image.transform(output_shape[::-1], Transform.AFFINE, data=pdata, resample=resample, fillcolor=cval)
 
     # params for scipy
     # swap X and Y axis for scipy
@@ -1332,7 +1333,7 @@ def extract_polygons(im: Image.Image,
                         for i in range(0, len(source_envelope)-3, 2)
                     ]
                     # warp
-                    resample = {0: Image.NEAREST, 1: Image.BILINEAR, 2: Image.BICUBIC, 3: Image.BICUBIC}.get(order, Image.NEAREST)
+                    resample = {0: Resampling.NEAREST, 1: Resampling.BILINEAR, 2: Resampling.BICUBIC, 3: Resampling.BICUBIC}.get(order, Resampling.NEAREST)
                     i = patch.transform((output_shape[1], output_shape[0]), Image.MESH, data=deform_mesh, resample=resample)
 
             yield i.crop(i.getbbox()), line

kraken/lib/train.py

@@ -406,9 +406,10 @@ def __init__(self,
         if format_type == 'binary':
             legacy_train_status = train_set.legacy_polygons_status
             if val_set and val_set.legacy_polygons_status != legacy_train_status:
-                logger.warning(
-                    f'Train and validation set have different legacy polygon status: {legacy_train_status} and {val_set.legacy_polygons_status}.'
-                     'Train set status prevails.')
+                logger.warning('Train and validation set have different legacy '
+                               f'polygon status: {legacy_train_status} and '
+                               f'{val_set.legacy_polygons_status}. Train set '
+                               'status prevails.')
             if legacy_train_status == "mixed":
                 logger.warning('Mixed legacy polygon status in training dataset. Consider recompilation.')
                 legacy_train_status = False

kraken/lib/vgsl.py

@@ -141,7 +141,7 @@ def __init__(self, spec: str) -> None:
                                               'one_channel_mode': None,
                                               'model_type': None,
                                               'hyper_params': {},
-                                              'legacy_polygons': False} # enable new polygons by default on new models
+                                              'legacy_polygons': False}  # enable new polygons by default on new models
         self._aux_layers = nn.ModuleDict()
 
         self.idx = -1
@@ -313,7 +313,7 @@ def _deserialize_layers(name, layer):
                                             'one_channel_mode': '1',
                                             'model_type': None,
                                             'hyper_params': {},
-                                            'legacy_polygons': True} # disable new polygons by default on load
+                                            'legacy_polygons': True}  # disable new polygons by default on load
 
         if 'kraken_meta' in mlmodel.user_defined_metadata:
             nn.user_metadata.update(json.loads(mlmodel.user_defined_metadata['kraken_meta']))
@@ -368,7 +368,7 @@ def aux_layers(self, val: Dict[str, torch.nn.Module]):
     @property
     def use_legacy_polygons(self):
         return self.user_metadata.get('legacy_polygons', True)
-    
+
     @use_legacy_polygons.setter
     def use_legacy_polygons(self, val: bool):
         self.user_metadata['legacy_polygons'] = val

tests/test_newpolygons.py

@@ -5,7 +5,7 @@
 import tempfile
 from unittest.mock import Mock, patch
 from pathlib import Path
-from traceback import print_exception
+from traceback import print_exc
 import warnings
 from typing import Optional, List, Union
 
@@ -89,7 +89,7 @@ def _test_krakencli(self, extractor_mock: Mock, *, args, force_no_legacy: bool=F
         print("kraken", *args)
 
         if result.exception:
-            print_exception(result.exception)
+            print_exc()
 
         self.assertEqual(result.exit_code, 0)
         extractor_mock.assert_called()
@@ -106,7 +106,7 @@ def _test_ketoscli(self, *, args, expect_legacy: bool, check_exit_code: Optional
             print("ketos", *args)
             if result.exception:
                 print(result.output)
-                print_exception(result.exception)
+                print_exc()
 
             if check_exit_code is not None:
                 if isinstance(check_exit_code, int):