orthoseg 0.8.0.dev0__tar.gz → 0.8.0.dev1__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (77) hide show
  1. {orthoseg-0.8.0.dev0/orthoseg.egg-info → orthoseg-0.8.0.dev1}/PKG-INFO +1 -1
  2. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/helpers/config_helper.py +108 -70
  3. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/lib/predicter.py +29 -10
  4. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/lib/trainer.py +39 -21
  5. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/load_images.py +2 -3
  6. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/load_sampleprojects.py +43 -23
  7. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/model/model_factory.py +164 -53
  8. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/model/model_helper.py +37 -15
  9. orthoseg-0.8.0.dev1/orthoseg/model/model_weights_helper.py +121 -0
  10. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/postprocess.py +2 -3
  11. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/predict.py +2 -3
  12. orthoseg-0.8.0.dev1/orthoseg/project_defaults.ini +841 -0
  13. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/train.py +5 -4
  14. orthoseg-0.8.0.dev1/orthoseg/util/_processing_util.py +218 -0
  15. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/config_util.py +7 -2
  16. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/general_util.py +0 -66
  17. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/image_util.py +114 -13
  18. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/log_util.py +5 -2
  19. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/validate.py +2 -3
  20. orthoseg-0.8.0.dev1/orthoseg/version.txt +1 -0
  21. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1/orthoseg.egg-info}/PKG-INFO +1 -1
  22. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg.egg-info/SOURCES.txt +6 -2
  23. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/pyproject.toml +3 -2
  24. orthoseg-0.8.0.dev0/tests/test_config.py → orthoseg-0.8.0.dev1/tests/test_config_helper.py +85 -2
  25. orthoseg-0.8.0.dev1/tests/test_config_util.py +31 -0
  26. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_end2end.py +66 -13
  27. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_helper.py +2 -2
  28. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_image_util.py +116 -24
  29. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_load_images.py +3 -3
  30. orthoseg-0.8.0.dev1/tests/test_load_sampleprojects.py +163 -0
  31. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_model_factory.py +84 -13
  32. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_model_helper.py +27 -3
  33. orthoseg-0.8.0.dev1/tests/test_model_weights_helper.py +58 -0
  34. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_postprocess.py +1 -1
  35. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_predict.py +2 -2
  36. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_train.py +2 -5
  37. orthoseg-0.8.0.dev1/tests/test_trainer.py +37 -0
  38. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_validate.py +8 -15
  39. orthoseg-0.8.0.dev0/orthoseg/project_defaults.ini +0 -524
  40. orthoseg-0.8.0.dev0/orthoseg/util/git_downloader.py +0 -150
  41. orthoseg-0.8.0.dev0/orthoseg/version.txt +0 -1
  42. orthoseg-0.8.0.dev0/tests/test_load_sampleprojects.py +0 -74
  43. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/LICENSE.MD +0 -0
  44. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/MANIFEST.in +0 -0
  45. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/README.md +0 -0
  46. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/__init__.py +0 -0
  47. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/_compat.py +0 -0
  48. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/helpers/__init__.py +0 -0
  49. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/helpers/email_helper.py +0 -0
  50. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/helpers/vectorfile_helper.py +0 -0
  51. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/lib/__init__.py +0 -0
  52. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/lib/cleanup.py +0 -0
  53. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/lib/postprocess_predictions.py +0 -0
  54. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/lib/prepare_traindatasets.py +0 -0
  55. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/model/__init__.py +0 -0
  56. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/py.typed +0 -0
  57. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/scriptrunner.py +0 -0
  58. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/scriptrunner_defaults.ini +0 -0
  59. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/__init__.py +0 -0
  60. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/data.py +0 -0
  61. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/progress_util.py +0 -0
  62. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg/util/vector_util.py +0 -0
  63. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg.egg-info/dependency_links.txt +0 -0
  64. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg.egg-info/entry_points.txt +0 -0
  65. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg.egg-info/requires.txt +0 -0
  66. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/orthoseg.egg-info/top_level.txt +0 -0
  67. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/setup.cfg +0 -0
  68. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/setup.py +0 -0
  69. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/__init__.py +0 -0
  70. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_cleanup.py +0 -0
  71. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_data.py +0 -0
  72. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_postprocess_predictions.py +0 -0
  73. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_predicter.py +0 -0
  74. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_prepare_traindata.py +0 -0
  75. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_vector_util.py +0 -0
  76. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_vectorfile_helper.py +0 -0
  77. {orthoseg-0.8.0.dev0 → orthoseg-0.8.0.dev1}/tests/test_version.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: orthoseg
3
- Version: 0.8.0.dev0
3
+ Version: 0.8.0.dev1
4
4
  Summary: Package to make it easier to segment orthophotos.
5
5
  Home-page: https://github.com/orthoseg/orthoseg
6
6
  Author: Pieter Roggemans
@@ -10,6 +10,7 @@ import pprint
10
10
  import re
11
11
  import shutil
12
12
  import tempfile
13
+ import warnings
13
14
  from pathlib import Path
14
15
  from typing import Any
15
16
 
@@ -19,8 +20,15 @@ from osgeo import gdal
19
20
 
20
21
  from orthoseg._compat import KERAS_GTE_3
21
22
  from orthoseg.lib.prepare_traindatasets import LabelInfo
23
+ from orthoseg.model.model_weights_helper import get_weights_types_for_architecture
22
24
  from orthoseg.util import config_util
23
- from orthoseg.util.image_util import FileLayerSource, WMSLayerSource, has_switched_axes
25
+ from orthoseg.util.image_util import (
26
+ FileLayerSource,
27
+ WMSLayerSource,
28
+ create_roi_for_dir,
29
+ create_vrt_for_dir,
30
+ has_switched_axes,
31
+ )
24
32
 
25
33
  # Get a logger...
26
34
  logger = logging.getLogger(__name__)
@@ -75,12 +83,14 @@ def read_orthoseg_config(config_path: Path, overrules: list[str] | None = None):
75
83
  config_path (Path): path to the configuration file to read.
76
84
  overrules (list[str], optional): list of config options that will overrule other
77
85
  ways to supply configuration. They should be specified as a list of
78
- "<section>.<parameter>=<value>" strings. Defaults to None.
86
+ "<section>.<key>=<value>" strings. Defaults to None.
79
87
  """
80
88
  # Set the temporary directory
81
89
  _set_tmp_dir()
82
90
 
83
91
  # Determine list of config files that should be loaded
92
+ # Support using ~ in the config path by expanding it
93
+ config_path = config_path.expanduser()
84
94
  config_paths = config_util.get_config_files(config_path)
85
95
 
86
96
  # If there are overrules, write them to a temporary configuration file.
@@ -149,7 +159,7 @@ def read_orthoseg_config(config_path: Path, overrules: list[str] | None = None):
149
159
  segment_subject = general.get("segment_subject")
150
160
  if segment_subject is None or segment_subject == "MUST_OVERRIDE":
151
161
  raise ValueError(
152
- "Projectconfig parameter general.segment_subject needs to be overruled to"
162
+ "Projectconfig parameter general.segment_subject needs to be overruled to "
153
163
  "a proper name in a specific project config file, \nwith config_filepaths "
154
164
  f"{config_paths}"
155
165
  )
@@ -172,7 +182,7 @@ def read_orthoseg_config(config_path: Path, overrules: list[str] | None = None):
172
182
  )
173
183
  dirs["projects_dir"] = projects_dir_absolute.as_posix()
174
184
 
175
- # Apply some defaults that need some logic.
185
+ # Some version-specific defaults
176
186
  if train.get("save_format") is None:
177
187
  train["save_format"] = "keras" if KERAS_GTE_3 else "h5"
178
188
  elif train.get("save_format") == "keras" and not KERAS_GTE_3:
@@ -186,6 +196,29 @@ def read_orthoseg_config(config_path: Path, overrules: list[str] | None = None):
186
196
  "Valid options are 'keras', 'h5', or 'tf'."
187
197
  )
188
198
 
199
+ if train.get("optimizer") is None:
200
+ if KERAS_GTE_3:
201
+ train["optimizer"] = "AdamW"
202
+ else:
203
+ # On keras 2, AdamW gives an error when training starts.
204
+ train["optimizer"] = "Adam"
205
+
206
+ # Some defaults that need some logic.
207
+ if train.get("weights_type") == "aerial_if_available":
208
+ # If aerial weights are available for the architecture, use them, otherwise use
209
+ # imagenet weights.
210
+ architecture = model.get("architecture")
211
+ weight_types = get_weights_types_for_architecture(architecture)
212
+ if len(weight_types) > 0 and "aerial" in weight_types:
213
+ train["weights_type"] = "aerial"
214
+ else:
215
+ warnings.warn(
216
+ f"No aerial pretrained weights available for {architecture=}, so "
217
+ f"falling back to imagenet pretrained weights.",
218
+ stacklevel=2,
219
+ )
220
+ train["weights_type"] = "imagenet"
221
+
189
222
  # Read the layer config
190
223
  layer_config_filepath = files.getpath("image_layers_config_filepath")
191
224
  global layer_config_filepath_used
@@ -329,44 +362,64 @@ def _read_layer_config(layer_config_filepath: Path) -> dict:
329
362
  switch_axes = has_switched_axes(crs)
330
363
  image_layers[image_layer]["switch_axes"] = switch_axes
331
364
 
365
+ # Read nb_concurrent calls param
366
+ image_layers[image_layer]["nb_concurrent_calls"] = layer_config[
367
+ image_layer
368
+ ].getint("nb_concurrent_calls", fallback=1)
369
+
370
+ # Check if a region of interest is specified as file or bbox
371
+ image_layers[image_layer]["roi_filepath"] = layer_config[image_layer].getpath(
372
+ "roi_filepath", fallback=None
373
+ )
374
+ bbox_tuple = None
375
+ if layer_config.has_option(image_layer, "bbox"):
376
+ bbox_list = layer_config[image_layer].getlist("bbox")
377
+ bbox_tuple = (
378
+ float(bbox_list[0]),
379
+ float(bbox_list[1]),
380
+ float(bbox_list[2]),
381
+ float(bbox_list[3]),
382
+ )
383
+ image_layers[image_layer]["bbox"] = bbox_tuple
384
+
385
+ # Check if the grid xmin and xmax are specified
386
+ image_layers[image_layer]["grid_xmin"] = layer_config[image_layer].getfloat(
387
+ "grid_xmin", fallback=0
388
+ )
389
+ image_layers[image_layer]["grid_ymin"] = layer_config[image_layer].getfloat(
390
+ "grid_ymin", fallback=0
391
+ )
392
+
393
+ # Check if a image_pixels_ignore_border is specified
394
+ image_layers[image_layer]["image_pixels_ignore_border"] = layer_config[
395
+ image_layer
396
+ ].getint("image_pixels_ignore_border", fallback=0)
397
+
398
+ # Convert pixel_x_size and pixel_y_size to float
399
+ image_layers[image_layer]["pixel_x_size"] = layer_config[image_layer].getfloat(
400
+ "pixel_x_size"
401
+ )
402
+ image_layers[image_layer]["pixel_y_size"] = layer_config[image_layer].getfloat(
403
+ "pixel_y_size"
404
+ )
405
+
332
406
  # If the layer source(s) are specified in a json parameter, parse it
333
407
  if "layersources" in image_layers[image_layer]:
334
408
  image_layers[image_layer]["layersources"] = layer_config[
335
409
  image_layer
336
410
  ].getdict("layersources")
337
411
  else:
338
- # If not, the layersource is specified in some top-level parameters
339
- layersource = {}
340
- layersource_keys = [
341
- "wms_server_url",
342
- "wms_version",
343
- "wms_layernames",
344
- "wms_layerstyles",
345
- "bands",
346
- "wms_username",
347
- "wms_password",
348
- "random_sleep",
349
- "wms_ignore_capabilities_url",
350
- "path",
351
- "layername",
352
- "wmts_server_url",
353
- "wmts_version",
354
- "wmts_layernames",
355
- "wmts_layerstyles",
356
- "wmts_tile_matrix_set",
357
- "wmts_xyz",
412
+ # If not, the layersource should be specified in the top-level parameters
413
+ image_layers[image_layer]["layersources"] = [
414
+ dict(layer_config[image_layer])
358
415
  ]
359
- for key in layersource_keys:
360
- if key in layer_config[image_layer]:
361
- layersource[key] = layer_config[image_layer][key]
362
- image_layers[image_layer]["layersources"] = [layersource]
363
416
 
364
417
  # Convert the layersource dicts to layersource objects
365
418
  layersource_objects = []
366
419
  for layersource in image_layers[image_layer]["layersources"]:
367
420
  layersource_object: WMSLayerSource | FileLayerSource
368
421
  try:
369
- # If not, the layersource should be specified in seperate parameters
422
+ # If not, the layersource should be specified in separate parameters
370
423
  if "wms_server_url" in layersource:
371
424
  layersource_object = WMSLayerSource(
372
425
  wms_server_url=layersource["wms_server_url"],
@@ -394,6 +447,32 @@ def _read_layer_config(layer_config_filepath: Path) -> dict:
394
447
  # Resolve relative path based on layer_config_filepath.parent
395
448
  path = layer_config_filepath.parent / layersource["path"]
396
449
  path = path.resolve()
450
+
451
+ # The path is a directory, so we need to create a VRT file for it.
452
+ if path.is_dir():
453
+ file_patterns = _str2list(layersource.get("file_patterns"))
454
+ if file_patterns is None:
455
+ raise ValueError(
456
+ f"file_patterns should be specified if path points to "
457
+ f"a directory {path}"
458
+ )
459
+ path = create_vrt_for_dir(
460
+ path, file_patterns, crs=layersource.get("projection")
461
+ )
462
+
463
+ # If no roi is specified, create one based on the dir contents.
464
+ if (
465
+ image_layers[image_layer].get("roi_filepath") is None
466
+ and image_layers[image_layer].get("bbox") is None
467
+ ):
468
+ image_layers[image_layer]["roi_filepath"] = (
469
+ create_roi_for_dir(
470
+ path.parent,
471
+ file_patterns,
472
+ crs=layersource.get("projection"),
473
+ )
474
+ )
475
+
397
476
  layersource_object = FileLayerSource(
398
477
  path=path,
399
478
  layernames=_str2list(layersource["layername"]),
@@ -411,47 +490,6 @@ def _read_layer_config(layer_config_filepath: Path) -> dict:
411
490
  layersource_objects.append(layersource_object)
412
491
  image_layers[image_layer]["layersources"] = layersource_objects
413
492
 
414
- # Read nb_concurrent calls param
415
- image_layers[image_layer]["nb_concurrent_calls"] = layer_config[
416
- image_layer
417
- ].getint("nb_concurrent_calls", fallback=1)
418
-
419
- # Check if a region of interest is specified as file or bbox
420
- image_layers[image_layer]["roi_filepath"] = layer_config[image_layer].getpath(
421
- "roi_filepath", fallback=None
422
- )
423
- bbox_tuple = None
424
- if layer_config.has_option(image_layer, "bbox"):
425
- bbox_list = layer_config[image_layer].getlist("bbox")
426
- bbox_tuple = (
427
- float(bbox_list[0]),
428
- float(bbox_list[1]),
429
- float(bbox_list[2]),
430
- float(bbox_list[3]),
431
- )
432
- image_layers[image_layer]["bbox"] = bbox_tuple
433
-
434
- # Check if the grid xmin and xmax are specified
435
- image_layers[image_layer]["grid_xmin"] = layer_config[image_layer].getfloat(
436
- "grid_xmin", fallback=0
437
- )
438
- image_layers[image_layer]["grid_ymin"] = layer_config[image_layer].getfloat(
439
- "grid_ymin", fallback=0
440
- )
441
-
442
- # Check if a image_pixels_ignore_border is specified
443
- image_layers[image_layer]["image_pixels_ignore_border"] = layer_config[
444
- image_layer
445
- ].getint("image_pixels_ignore_border", fallback=0)
446
-
447
- # Convert pixel_x_size and pixel_y_size to float
448
- image_layers[image_layer]["pixel_x_size"] = layer_config[image_layer].getfloat(
449
- "pixel_x_size"
450
- )
451
- image_layers[image_layer]["pixel_y_size"] = layer_config[image_layer].getfloat(
452
- "pixel_y_size"
453
- )
454
-
455
493
  return image_layers
456
494
 
457
495
 
@@ -25,7 +25,7 @@ import rasterio.plot as rio_plot
25
25
  import tensorflow as tf
26
26
 
27
27
  import orthoseg.lib.postprocess_predictions as postp
28
- from orthoseg.util import general_util, image_util
28
+ from orthoseg.util import _processing_util, image_util
29
29
  from orthoseg.util.progress_util import ProgressLogger
30
30
 
31
31
  # Get a logger...
@@ -456,7 +456,7 @@ def _predict_layer(
456
456
  # We don't want the postprocess workers to block the entire system,
457
457
  # so make them a bit nicer
458
458
  def init_postprocess_worker():
459
- general_util.setprocessnice(15)
459
+ _processing_util.setprocessnice(15)
460
460
 
461
461
  with (
462
462
  futures.ThreadPoolExecutor(nb_parallel_read) as read_pool,
@@ -575,6 +575,11 @@ def _predict_layer(
575
575
  }
576
576
  )
577
577
 
578
+ except Exception as ex: # pragma: no cover
579
+ nb_errors += 1
580
+ image_path = read_queue[future]
581
+ _handle_error(image_path, ex, images_error_log_filepath)
582
+
578
583
  finally:
579
584
  # Remove from queue...
580
585
  del read_queue[future]
@@ -695,11 +700,7 @@ def _predict_layer(
695
700
 
696
701
  if output_vector_path is None:
697
702
  # No vector output, so we are ready with this image
698
- with images_done_log_filepath.open(
699
- "a+"
700
- ) as image_donelog_file:
701
- image_donelog_file.write(f"{image_path.name}\n")
702
-
703
+ _write_to_done_log(image_path, images_done_log_filepath)
703
704
  nb_done += 1
704
705
  else:
705
706
  # Schedule `_write_vector_result` to move the result of the
@@ -844,8 +845,26 @@ def _write_vector_result(
844
845
  gfo.remove(partial_vector_path)
845
846
 
846
847
  # Write filepath to file with files that are done
847
- with images_done_log_filepath.open("a+") as image_donelog_file:
848
- image_donelog_file.write(f"{image_path.name}\n")
848
+ _write_to_done_log(image_path, images_done_log_filepath)
849
+
850
+
851
+ def _write_to_done_log(image_path: Path, images_done_log_filepath: Path):
852
+ nb_tries = 0
853
+ while True:
854
+ try:
855
+ with images_done_log_filepath.open("a+") as image_donelog_file:
856
+ image_donelog_file.write(f"{image_path.name}\n")
857
+ break
858
+
859
+ except Exception as ex: # pragma: no cover
860
+ nb_tries += 1
861
+ time.sleep(0.1)
862
+
863
+ if nb_tries >= 5:
864
+ message = (
865
+ f"Write to done log failed {nb_tries} times for {image_path}: {ex}"
866
+ )
867
+ raise RuntimeError(message) from ex
849
868
 
850
869
 
851
870
  def _handle_error(image_path: Path, ex: Exception, log_path: Path):
@@ -905,7 +924,7 @@ def read_image(
905
924
 
906
925
  # Read worked, so jump out of the loop...
907
926
  break
908
- except RuntimeError as ex: # pragma: no cover
927
+ except Exception as ex: # pragma: no cover
909
928
  retry_count += 1
910
929
  logger.warning(f"Read failed, retry nb {retry_count} for {image_path}")
911
930
  if retry_count >= 3:
@@ -26,6 +26,7 @@ def train(
26
26
  segment_subject: str,
27
27
  traindata_id: int,
28
28
  hyperparams: mh.HyperParams,
29
+ weights_dir: Path | None = None,
29
30
  model_preload_filepath: Path | None = None,
30
31
  image_width: int = 512,
31
32
  image_height: int = 512,
@@ -53,6 +54,7 @@ def train(
53
54
  segment_subject (str): segment subject
54
55
  traindata_id (int): train data version
55
56
  hyperparams (mh.HyperParams): the hyper parameters to use for the model
57
+ weights_dir: directory where pretrained weights are cached and read from
56
58
  model_preload_filepath: filepath to the model to continue training on,
57
59
  or None if you want to start from scratch
58
60
  image_width: width the input images will be rescaled to for training
@@ -64,25 +66,38 @@ def train(
64
66
  save_augmented_subdir: (str, optional): the subdirectory to save the augmented
65
67
  images to. If None, the images aren't saved. Defaults to None.
66
68
  """
67
- # Get the max epoch number from log file if it exists...
68
- start_epoch = 0
69
+ # First if there exists a model already for the traindata_id,...
70
+ curr_best_model = mh.get_best_model(
71
+ model_dir=model_save_dir,
72
+ segment_subject=segment_subject,
73
+ traindata_id=traindata_id,
74
+ architecture_id=hyperparams.architecture.architecture_id,
75
+ trainparams_id=hyperparams.train.trainparams_id,
76
+ )
69
77
  model_save_base_filename = mh.format_model_basefilename(
70
78
  segment_subject=segment_subject,
71
79
  traindata_id=traindata_id,
72
80
  architecture_id=hyperparams.architecture.architecture_id,
73
81
  trainparams_id=hyperparams.train.trainparams_id,
74
82
  )
75
- csv_log_filepath = model_save_dir / (model_save_base_filename + "_log.csv")
76
- if csv_log_filepath.exists() and csv_log_filepath.stat().st_size > 0:
83
+
84
+ # If there is an existing model and we want to continue training on that, recuperate
85
+ # the epoch.
86
+ start_epoch = 0
87
+ csv_log_path = model_save_dir / f"{model_save_base_filename}_train_log.csv"
88
+ model_json_path = model_save_dir / f"{model_save_base_filename}_model.json"
89
+ hyperparams_path = model_save_dir / f"{model_save_base_filename}_hyperparams.json"
90
+
91
+ if curr_best_model is not None:
77
92
  if not model_preload_filepath:
78
93
  message = (
79
- "log file exists but preload model file not specified: "
80
- f"{csv_log_filepath}"
94
+ "Model exists but preload model file not specified: "
95
+ f"{curr_best_model['model_filepath']}"
81
96
  )
82
97
  logger.critical(message)
83
98
  raise ValueError(message)
84
99
 
85
- train_log_df = pd.read_csv(csv_log_filepath, sep=";", usecols=["epoch", "lr"])
100
+ train_log_df = pd.read_csv(csv_log_path, sep=";", usecols=["epoch", "lr"])
86
101
  assert isinstance(train_log_df, pd.DataFrame)
87
102
  logger.debug(f"train_log csv contents:\n{train_log_df}")
88
103
  start_epoch = train_log_df["epoch"].max()
@@ -94,6 +109,12 @@ def train(
94
109
  f"{hyperparams.train.optimizer_params['learning_rate']}"
95
110
  )
96
111
 
112
+ if start_epoch == 0:
113
+ # No existing model, so clean any existing files that might exist
114
+ csv_log_path.unlink(missing_ok=True)
115
+ model_json_path.unlink(missing_ok=True)
116
+ hyperparams_path.unlink(missing_ok=True)
117
+
97
118
  # If no existing model provided, create it from scratch
98
119
  if not model_preload_filepath:
99
120
  # If the encode should be frozen for some epochs...
@@ -108,16 +129,16 @@ def train(
108
129
  nb_channels=hyperparams.architecture.nb_channels,
109
130
  nb_classes=len(hyperparams.architecture.classes),
110
131
  activation=hyperparams.architecture.activation_function,
132
+ weights_type=hyperparams.train.weights_type,
133
+ weights_dir=weights_dir,
111
134
  freeze=freeze,
112
135
  )
113
136
 
114
137
  # Save the model architecture to json
115
- model_json_filepath = model_save_dir / f"{model_save_base_filename}_model.json"
116
- if not model_save_dir.exists():
117
- model_save_dir.mkdir(parents=True)
118
- if not model_json_filepath.exists():
119
- with model_json_filepath.open("w") as dst:
120
- dst.write(str(model.to_json()))
138
+ model_save_dir.mkdir(parents=True, exist_ok=True)
139
+ with model_json_path.open("w") as dst:
140
+ dst.write(str(model.to_json()))
141
+
121
142
  else:
122
143
  # If a preload model is provided, load that if it exists...
123
144
  if not model_preload_filepath.exists():
@@ -258,7 +279,7 @@ def train(
258
279
  train_callbacks.append(tensorboard_logger)
259
280
  if hyperparams.train.log_csv is True:
260
281
  csv_logger = keras.callbacks.CSVLogger(
261
- str(csv_log_filepath), append=True, separator=";"
282
+ str(csv_log_path), append=True, separator=";"
262
283
  )
263
284
  train_callbacks.append(csv_logger)
264
285
 
@@ -305,14 +326,11 @@ def train(
305
326
  )
306
327
 
307
328
  logger.info(f"{hyperparams.toJSON()}")
308
- hyperparams_filepath = (
309
- model_save_dir / f"{model_save_base_filename}_hyperparams.json"
310
- )
311
- hyperparams_filepath.write_text(hyperparams.toJSON())
329
+ hyperparams_path.write_text(hyperparams.toJSON())
312
330
 
313
331
  try:
314
- # If the encoder should be frozen for the first epochs, do so
315
332
  if hyperparams.train.nb_epoch_with_freeze > 0:
333
+ # First train only the top layers for a few epochs.
316
334
  logger.info(
317
335
  f"First train for {hyperparams.train.nb_epoch_with_freeze} epochs with "
318
336
  "frozen layers"
@@ -352,7 +370,7 @@ def train(
352
370
 
353
371
  # Write some reporting
354
372
  train_report_path = model_save_dir / (model_save_base_filename + "_report.pdf")
355
- train_log_df = pd.read_csv(csv_log_filepath, sep=";")
373
+ train_log_df = pd.read_csv(csv_log_path, sep=";")
356
374
  assert isinstance(train_log_df, pd.DataFrame)
357
375
  columns_to_keep = []
358
376
  for column in train_log_df.columns:
@@ -362,7 +380,7 @@ def train(
362
380
  train_log_vis_df = train_log_df[columns_to_keep]
363
381
  fig = train_log_vis_df.plot().get_figure()
364
382
  if fig is not None:
365
- fig.savefig(str(train_report_path))
383
+ fig.savefig(str(train_report_path)) # type: ignore[union-attr]
366
384
 
367
385
  finally:
368
386
  # Release the memory from the GPU...
@@ -40,8 +40,7 @@ def _load_images_args(args) -> argparse.Namespace:
40
40
  "config_overrules",
41
41
  nargs="*",
42
42
  help=(
43
- "Supply any number of config overrules like this: "
44
- "<section>.<parameter>=<value>"
43
+ "Supply any number of config overrules like this: <section>.<key>=<value>"
45
44
  ),
46
45
  )
47
46
 
@@ -61,7 +60,7 @@ def load_images(
61
60
  images. Defaults to False.
62
61
  config_overrules (list[str], optional): list of config options that will
63
62
  overrule other ways to supply configuration. They should be specified in the
64
- form of "<section>.<parameter>=<value>". Defaults to None.
63
+ form of "<section>.<key>=<value>". Defaults to None.
65
64
  """
66
65
  # Init
67
66
  # Load the config and save in a bunch of global variables zo it
@@ -2,13 +2,15 @@
2
2
 
3
3
  import argparse
4
4
  import logging
5
+ import shutil
5
6
  import sys
7
+ import tempfile
8
+ import urllib.request
9
+ import zipfile
6
10
  from pathlib import Path
7
11
 
8
12
  import gdown
9
13
 
10
- from orthoseg.util import git_downloader
11
-
12
14
  # Get a logger...
13
15
  logger = logging.getLogger(__name__)
14
16
 
@@ -59,43 +61,61 @@ def load_sampleprojects(dest_dir: Path, ssl_verify: bool | str = True):
59
61
  this will be used. In corporate networks using a proxy server this is often
60
62
  needed to avoid CERTIFICATE_VERIFY_FAILED errors. Defaults to True.
61
63
  """
64
+ dest_dir = dest_dir.expanduser()
62
65
  dest_dir_full = dest_dir / "sample_projects"
63
66
  if dest_dir_full.exists():
64
67
  raise ValueError(f"Destination directory already exists: {dest_dir_full}")
65
68
 
66
69
  # Download
67
70
  print(f"Start download of sample projects to {dest_dir_full!s}")
68
- git_downloader.download(
69
- repo_url="https://github.com/orthoseg/orthoseg/tree/main/sample_projects",
70
- output_dir=dest_dir,
71
- ssl_verify=ssl_verify,
72
- )
71
+ with tempfile.TemporaryDirectory(ignore_cleanup_errors=True) as tmp_dir:
72
+ # Download the sample projects to a temporary directory first
73
+ url = "https://github.com/orthoseg/orthoseg_sample_projects/archive/refs/tags/v0.1.0.zip"
74
+ tmp_zip_path = Path(tmp_dir) / "sample_projects.zip"
75
+ urllib.request.urlretrieve(url, tmp_zip_path)
76
+
77
+ # Unzip the downloaded file
78
+ tmp_proj_dir = Path(tmp_dir) / "sample_projects_tmp"
79
+ with zipfile.ZipFile(tmp_zip_path, "r") as zip_file:
80
+ zip_file.extractall(tmp_proj_dir)
81
+ # Get the single subdirectory in the unzipped folder
82
+ subdirs = [d for d in tmp_proj_dir.iterdir() if d.is_dir()]
83
+ if len(subdirs) != 1:
84
+ raise ValueError(
85
+ f"Expected exactly one subdirectory, but found {len(subdirs)}"
86
+ )
87
+
88
+ # Move the unzipped sample projects to the destination directory
89
+ dest_dir.mkdir(parents=True, exist_ok=True)
90
+ shutil.move(subdirs[0] / "sample_projects", dest_dir)
91
+
73
92
  print("Download finished")
93
+
74
94
  print("Start download of footballfields pretrained neural net")
75
- footballfields_model_dir = dest_dir_full / "footballfields/models"
76
- footballfields_model_dir.mkdir(parents=True, exist_ok=True)
77
- model_hdf5_path = footballfields_model_dir / "footballfields_01_0.92512_242.hdf5"
78
- if model_hdf5_path.exists() is False:
95
+ verify = True if ssl_verify is None else ssl_verify
96
+ model_dir = dest_dir_full / "footballfields/models"
97
+ model_dir.mkdir(parents=True, exist_ok=True)
98
+
99
+ model_hdf5_path = model_dir / "footballfields_01_0.97392_201.hdf5"
100
+ if not model_hdf5_path.exists():
79
101
  gdown.download(
80
- id="1XmAenCW6K_RVwqC6xbkapJ5ws-f7-QgH",
102
+ id="1UlNorZ74ADCr3pL4MCJ_tnKRNoeZX79g",
81
103
  output=str(model_hdf5_path),
82
- verify=ssl_verify,
104
+ verify=verify,
83
105
  )
84
- model_hyperparams_path = (
85
- footballfields_model_dir / "footballfields_01_hyperparams.json"
86
- )
87
- if model_hyperparams_path.exists() is False:
106
+ model_hyperparams_path = model_dir / "footballfields_01_hyperparams.json"
107
+ if not model_hyperparams_path.exists():
88
108
  gdown.download(
89
- id="1umxcd4RkB81sem9PdIpLoWeiIW8ga1u7",
109
+ id="1NwrVVjx9IsjvaioQ4-bkPMrq7S6HeWIo",
90
110
  output=str(model_hyperparams_path),
91
- verify=ssl_verify,
111
+ verify=verify,
92
112
  )
93
- model_modeljson_path = footballfields_model_dir / "footballfields_01_model.json"
94
- if model_modeljson_path.exists() is False:
113
+ model_modeljson_path = model_dir / "footballfields_01_model.json"
114
+ if not model_modeljson_path.exists():
95
115
  gdown.download(
96
- id="16qe8thBTrO3dFfLMU1T22gWcfHVXt8zQ",
116
+ id="1LNPLypM5in3aZngBKK_U4Si47Oe97ZWN",
97
117
  output=str(model_modeljson_path),
98
- verify=ssl_verify,
118
+ verify=verify,
99
119
  )
100
120
  print("Download finished")
101
121