INFO:tensorflow:Caught OutOfRangeError. Stopping Training. Graph execution error:
Detected at node 'batch_1' defined at (most recent call last):
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 487, in <module>
tf.app.run()
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 308, in run
_run_main(main, args)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 254, in _run_main
sys.exit(main(argv))
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 296, in main
samples_val = input_generator.get(
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\utils\input_generator.py", line 162, in get
return tf.train.batch(
Node: 'batch_1'
Detected at node 'batch_1' defined at (most recent call last):
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 487, in <module>
tf.app.run()
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 308, in run
_run_main(main, args)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 254, in _run_main
sys.exit(main(argv))
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 296, in main
samples_val = input_generator.get(
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\utils\input_generator.py", line 162, in get
return tf.train.batch(
Node: 'batch_1'
2 root error(s) found.
(0) OUT_OF_RANGE: PaddingFIFOQueue '_5_batch_1/padding_fifo_queue' is closed and has insufficient elements (requested 18, current size 0)
[[{{node batch_1}}]]
[[batch_1/_2171]]
(1) OUT_OF_RANGE: PaddingFIFOQueue '_5_batch_1/padding_fifo_queue' is closed and has insufficient elements (requested 18, current size 0)
[[{{node batch_1}}]]
0 successful operations.
0 derived errors ignored.
Original stack trace for 'batch_1':
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 487, in <module>
tf.app.run()
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\platform\app.py", line 36, in run
_run(main=main, argv=argv, flags_parser=_parse_flags_tolerate_undef)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 308, in run
_run_main(main, args)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 254, in _run_main
sys.exit(main(argv))
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 296, in main
samples_val = input_generator.get(
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\utils\input_generator.py", line 162, in get
return tf.train.batch(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\util\deprecation.py", line 357, in new_func
return func(*args, **kwargs)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\training\input.py", line 1006, in batch
return _batch(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\training\input.py", line 784, in _batch
dequeued = queue.dequeue_up_to(batch_size, name=name)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\ops\data_flow_ops.py", line 533, in dequeue_up_to
ret = gen_data_flow_ops.queue_dequeue_up_to_v2(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\ops\gen_data_flow_ops.py", line 3737, in queue_dequeue_up_to_v2
_, _, _op, _outputs = _op_def_library._apply_op_helper(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 797, in _apply_op_helper
op = g._create_op_internal(op_type_name, inputs, dtypes=None,
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\framework\ops.py", line 3800, in _create_op_internal
ret = Operation(
I1226 02:16:37.282644 24784 learning.py:775] Caught OutOfRangeError. Stopping Training. Graph execution error:
Detected at node 'batch_1' defined at (most recent call last):
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 487, in <module>
tf.app.run()
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 308, in run
_run_main(main, args)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 254, in _run_main
sys.exit(main(argv))
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 296, in main
samples_val = input_generator.get(
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\utils\input_generator.py", line 162, in get
return tf.train.batch(
Node: 'batch_1'
Detected at node 'batch_1' defined at (most recent call last):
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 487, in <module>
tf.app.run()
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 308, in run
_run_main(main, args)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 254, in _run_main
sys.exit(main(argv))
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 296, in main
samples_val = input_generator.get(
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\utils\input_generator.py", line 162, in get
return tf.train.batch(
Node: 'batch_1'
2 root error(s) found.
(0) OUT_OF_RANGE: PaddingFIFOQueue '_5_batch_1/padding_fifo_queue' is closed and has insufficient elements (requested 18, current size 0)
[[{{node batch_1}}]]
[[batch_1/_2171]]
(1) OUT_OF_RANGE: PaddingFIFOQueue '_5_batch_1/padding_fifo_queue' is closed and has insufficient elements (requested 18, current size 0)
[[{{node batch_1}}]]
0 successful operations.
0 derived errors ignored.
Original stack trace for 'batch_1':
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 487, in <module>
tf.app.run()
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\platform\app.py", line 36, in run
_run(main=main, argv=argv, flags_parser=_parse_flags_tolerate_undef)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 308, in run
_run_main(main, args)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\absl\app.py", line 254, in _run_main
sys.exit(main(argv))
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\train.py", line 296, in main
samples_val = input_generator.get(
File "Z:\IdeaRepos\masterarbeit\Segmentation_Models\deeplab\utils\input_generator.py", line 162, in get
return tf.train.batch(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\util\deprecation.py", line 357, in new_func
return func(*args, **kwargs)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\training\input.py", line 1006, in batch
return _batch(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\training\input.py", line 784, in _batch
dequeued = queue.dequeue_up_to(batch_size, name=name)
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\ops\data_flow_ops.py", line 533, in dequeue_up_to
ret = gen_data_flow_ops.queue_dequeue_up_to_v2(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\ops\gen_data_flow_ops.py", line 3737, in queue_dequeue_up_to_v2
_, _, _op, _outputs = _op_def_library._apply_op_helper(
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 797, in _apply_op_helper
op = g._create_op_internal(op_type_name, inputs, dtypes=None,
File "C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\framework\ops.py", line 3800, in _create_op_internal
ret = Operation(
INFO:tensorflow:Finished training! Saving model to disk.
I1226 02:16:37.289651 24784 learning.py:777] Finished training! Saving model to disk.
C:\Users\Fistus\miniconda3\envs\code\lib\site-packages\tensorflow\python\summary\writer\writer.py:438: UserWarning: Attempting to use a closed FileWriter. The operation will be a noop unless the FileWriter is explicitly reopened.
warnings.warn("Attempting to use a closed FileWriter. "
Process finished with exit code 0
Accuracy: done
IoU: done (don't forget to run update tensor, not just iou - otherwise it will stay 0 forever)
Adding just tensors to summary like this did not work:
accuracy_validation = slim.metrics.accuracy(tf.to_int32(predictions_val),
tf.to_int32(labels_val), name="accuracy")
iou,conf_mat = slim.metrics.streaming_mean_iou(tf.to_int32(predictions_val),tf.to_int32(labels_val), num_classes=5, name="accuracy")
# Merge all summaries together.
summaries |= set(
tf.get_collection(tf.GraphKeys.SUMMARIES, first_clone_scope))
summary_acc = tf.summary.tensor_summary('val_acc', accuracy_validation)
summary_iou = tf.summary.tensor_summary('val_iou', iou)
summaries.add(summary_acc)
summaries.add(summary_iou)
summary_op = tf.summary.merge(list(summaries))W1223 15:04:45.235898 22556 plugin_event_accumulator.py:386] This summary with tag 'val_acc' is oddly not associated with a plugin.
W1223 15:04:45.240903 22556 plugin_event_accumulator.py:386] This summary with tag 'val_iou' is oddly not associated with a plugin.
One needs to run the tensors in the step function and add not the tensors but tf.scalar which are executed later:
accuracy_validation = slim.metrics.accuracy(tf.to_int32(predictions_val),
tf.to_int32(labels_val), name="accuracy")
iou, iou_update_op = slim.metrics.streaming_mean_iou(tf.to_int32(predictions_val),tf.to_int32(labels_val), num_classes=5, name="iou")
summaries.add(tf.summary.scalar('accuracy', accuracy_validation))
summaries.add(tf.summary.scalar('iou', iou))If the error error: Can't find libdevice directory ${CUDA_DIR}/nvvm/libdevice occurs:
Export the environment var XLA_FLAGS pointing to:
--xla_gpu_cuda_data_dir=C:\Users\Fistus\miniconda3\envs\code\Library
where this directory contains the path nvvm/libdevice (copied from cuda directory)!
Don't forget to restart Pycharm after creating the environment variable!
--logtostderr
--training_number_of_steps=90000
--train_split="train"
--model_variant="mobilenet_v2"
--output_stride=16
--decoder_output_stride=4
--train_crop_size=600
--train_crop_size=600
--train_batch_size=1
--initialize_last_layer=False
--last_layers_contain_logits_only=True
--train_batch_size=8
--dataset="de_top15"
--fine_tune_batch_norm=False
--train_logdir="K:/!!!!slurm trained models/test_train_mobilenet"
--dataset_dir="K:/FINAL_DATASETS/deeplab_de_top15_2"
--tf_initial_checkpoint="K:\!!!!slurm trained models\Pretrained weights\deeplab\deeplabv3_mnv2_cityscapes_train/model.ckpt"
Use this to train on eutop25 without residential area but with weights:
--logtostderr
--training_number_of_steps=900000
--train_split="train"
--model_variant="mobilenet_v2"
--output_stride=16
--decoder_output_stride=4
--train_crop_size=600
--train_crop_size=600
--initialize_last_layer=False
--save_summaries_images=True
--last_layers_contain_logits_only=True
--train_batch_size=4
--dataset="eu_top25_nores"
--fine_tune_batch_norm=False
--base_learning_rate=0.0001
--validation_check=500
--classweights=0.303529
--classweights=1.000000
--classweights=0.604396
--classweights=5.941638
--classweights=1.305352
--train_logdir="K:/!!!!slurm trained models/test_train_mobilenet_eu_classweights"
--dataset_dir="K:\nores_datasets\deeplab_eu_top25_cropped_nores/tfrecord"
--tf_initial_checkpoint="K:\!!!!slurm trained models\Pretrained weights\deeplab\deeplabv3_mnv2_cityscapes_train\model.ckpt"
Use tensorboard --logdir .\test_train_mobilenet\ --host 0.0.0.0 to monitor training.
Go to Url to access Tensorboard dashboard: http://localhost:6006/
-
13 größte städte in dtl. (nach berlin, ist ja schon gedownloaded) hinzugefügt
-
(y)Hamburg --- 663 sq km
-
(y)Muenchen --- 1015 sq km *neu
-
(y)koeln --- 615 sq km *neu
-
(y)Frankfurth am Main --- 206 sq km
-
(y)Stuttgart --- 256 sq km
-
(y)Düsseldorf --- 292 sq km
-
(y)Dortmund --- 554 sq km
-
(y)Essen --- 173 sq km
-
(y)Leipzig --- 268 sq km
-
(y)Bremen --- 400 sq km
-
(y)Dresden --- 439 sq km
-
(y)Hannover --- 452 sq km
-
(y)Nürnberg --- 182 sq km
Next: Alle importieren in postgis Coordinaten der bbox Aufschreiben und in Python übernehmen (DONE)
- (y)Hamburg --- 663 sq km
- (y)Muenchen --- 1015 sq km
- (y)Koeln --- 615 sq km
- (y)Frankfurth am Main --- 206 sq km
- (y)Stuttgart --- 256 sq km
- (y)Duesseldorf --- 292 sq km
- (y)Dortmund --- 554 sq km
- (y)Essen --- 173 sq km
- (y)Leipzig --- 268 sq km
- (y)Bremen --- 400 sq km
- (y)Dresden --- 439 sq km
- (y)Hannover --- 452 sq km
- (y)Nürnberg --- 182 sq km
Next: coords in python eintragen (DONE)
Downloadprozedur für neues dataset überlegen Problem: Nur 25k images von google am Tag möglich Ein Image mit 600x600px 17 zoom hat circa 437x437 meter (0.43 sq km) in auf deutschem/europäischem Breitengrad
Ausgerechnet mittels:
lat = 52.4983678856
lon = 13.3306850560
centerPoint = helper.G_LatLng(lat, lon)
corners = helper.getCorners(centerPoint, 17, 600, 600)
bbox = corners['W'], corners['S'], corners['E'], corners['N']
dis1 = helper.getMeterDistance(lon1=bbox[0], lat1=bbox[1], lon2=bbox[0], lat2=bbox[3])
dis2 = helper.getMeterDistance(lon1=bbox[0], lat1=bbox[1], lon2=bbox[2], lat2=bbox[1])
print(dis1)
print(dis2)
- Nicht mehr als 500 Bilder pro Stadt
- Dataset mit 7000 Bilden (incl. Berlin) erzeugt 'de_top14'
- Feature Distribution:
"building": 14.39
"road": 11.11
"unlabelled": 15.29
"water": 2.24
"residential": 32.44
"wood": 24.53
- Istanbul 8933 sq km y warning: 50% < unusable sea water
- Moskau 8846 sq km y
- London 2286 sq km y
- St. Petersburg 4212 sq km y
- Madrid 2749 sq km y
- Kiew 2340 sq km y
- Rom 1032 sq km y
- Paris 1070 sq km y could not load bigger bbox 'too many nodes error from hotosm'
- Minsk 1464 sq km y
- Wien 2079 sq km y
- Bukarest 1592 sq km y
- Budapest 1423 sq km y
- Warschau 1445 sq km y
- Barcelona 1350 sq km y
- Charkiv 3240 sq km y
- Mailand 2202 sq km y
- Prag 1547 sq km y
- Nischni Nowgorod 1092 sq km y
- Sofia 968 sq km y
- Belgrad 3579 sq km y
- Kasan 1743 sq km y
- Samara 956 sq km y
- Rostow am Don 1309 sq km y
- Birmingham 1588 sq km y
- Ufa 888 sq km y
-
Wichtig: Unterschiedliche viele Bilder von den Städten laden, sodass Fläche optimal ausgenutzt wird
-
Problem: Manche Städte haben sehr wenige gelabelte Gebäude (oft Viertel mit nur residential area ohne Gebäude)
-
Idee: Bilde für jede Stadt Quotient aus Gebäude(Anzahl)/Qudratkilometer und lasse diesen Wert in die Wahl der zu ladenden Gebäude für die Städte miteinfließen. Schließe ggf. zu schlechte Städte direkt aus
-
Big Todo: Rolle keine zufälligen Kartenausschnitte mehr sondern zerhacke das Bild von links nach rechts und oben nach unten. Dies könnte lange Wartezeiten beim Laden verursachen da oft große Gebiete mit unlabelt Data da sind. Schwierig ist die Implementierung da die BBox des Bildes erst aus der Center-Coordinate errechnet werden muss und von vornherein nicht klar ist "wie weit" man nach rechts bzw. nach unten gehen muss damit man das letzte Bild gerade nicht schneidet aber trotzdem nicht zuweit weg geht. Generische Lösung kompliziert wegen Erdkrümmung.
Außerdem auffällig:
- Mehrspurige Autobahnen sind oft zu dünn
- Schräge Aufnahme von Gebäuden macht Labelling unpräzise
- Meersegmente problematisch, da sie ungelabellet sind (Wird vermutlich Wasserklassifikation stark schwächen wenn zuviel davon da ist)
- Manche Gebäude sind einfach blau angemalt (Google Feature? Geheimgebäude?)
| Stadt | qm | Gebäude | Gebäude/qm | Anzahl Bilder |
|---|---|---|---|---|
| Istanbul | 8933 | 247320 | 27.686 | 1200 |
| Moskau | 8846 | 472169 | 53.376 | 1800 |
| London | 2286 | 400430 | 175.16 | 1000 |
| St. Petersburg | 4212 | 194267 | 46.122 | 1400 |
| Madrid | 2749 | 181060 | 65.863 | 1000 |
| Kiew | 2340 | 100724 | 43.044 | 700 |
| Rom | 1032 | 156758 | 151.89 | 450 |
| Paris | 1070 | 727437 | 679.84 | 1000 |
| Minsk | 1464 | 113943 | 77.82 | 400 |
| Wien | 2079 | 369930 | 177.93 | 800 |
| Bukarest | 1592 | 89771 | 56.388 | 400 |
| Budapest | 1423 | 205994 | 144.76 | 500 |
| Warschau | 1445 | 238325 | 164.93 | 500 |
| Barcelona | 1350 | 60619 | 44.90 | 250 |
| Charkiv | 3240 | 126490 | 37.871 | 700 |
| Mailand | 2202 | 224641 | 102.01 | 700 |
| Prag | 1547 | 318561 | 205.92 | 500 |
| Nischni Nowgorod | 1092 | 85291 | 78.105 | 400 |
| Sofia | 968 | 96911 | 100.11 | 380 |
| Belgrad | 3579 | 205518 | 57.423 | 500 |
| Kasan | 1743 | 71304 | 40.908 | 300 |
| Samara | 956 | 35416 | 37.046 | 300 |
| Rostow am Don | 1309 | 125197 | 95.643 | 400 |
| Birmingham | 1588 | 600898 | 378.39 | 700 |
| Ufa | 888 | 24986 | 28.137 | 200 |
Gesamt: 15580
Ermittlung günstiger Anzahl der Bilder pro Region (Grobe Abschätzungshilfe): Für Städte mit mehr als 30 Gebäuden pro qm:
(qm_nutzbar / 0.5 qm ) * 0.3
Für andere Städte :
(qm_nutzbar / 0.5 qm ) * 0.2
Images which are to similiar to each other shall be removed (that is, from a given image pair (im_1, im_2) with dist(im_1,im_2)<Threshold, im_2 will be removed from the dataset).
With a threshold of 0.0045, this would affect the dataset as it follows:
- de_top14: Reduced from 7000 to 6562 ~6.26% images lost
- de_top25_exde: Reduced from 16479 to 14601 ~11.4% images lost
- world_tiny2k: Reduced from 1999 to 1840 ~7.95% images lost
The training and validation sets are used during training.
for each epoch
for each training data instance
propagate error through the network
adjust the weights
calculate the accuracy over training data
for each validation data instance
calculate the accuracy over the validation data
if the threshold validation accuracy is met
exit training
else
continue training
Once you're finished training, then you run against your testing set and verify that the accuracy is sufficient.
Training Set: this data set is used to adjust the weights on the neural network.
Validation Set: this data set is used to minimize overfitting. You're not adjusting the weights of the network with this data set, you're just verifying that any increase in accuracy over the training data set actually yields an increase in accuracy over a data set that has not been shown to the network before, or at least the network hasn't trained on it (i.e. validation data set). If the accuracy over the training data set increases, but the accuracy over then validation data set stays the same or decreases, then you're overfitting your neural network and you should stop training.
Testing Set: this data set is used only for testing the final solution in order to confirm the actual predictive power of the network.
Different 5-Fold splits are checked and the most balanced ones chosen for 5-Fold-Cross-Validation
0: ['koel', 'dres', 'esse'] 1: ['duis', 'stut', 'hann'] 2: ['leip', 'muni', 'dort'] 3: ['brem', 'berl', 'fran'] 4: ['hamb', 'nuer', 'dues']
- [1408, 1401, 1414, 1405, 1404]
- Best fold splitting with std 4
e1: {'wood': 0.2983977884311869, 'road': 0.10457110755997469, 'residential': 0.3067120837279039, 'unlabelled': 0.1469938506155305, 'building': 0.11831597024936884, 'water': 0.02500919941603535}
e2:{'wood': 0.23019293956697626, 'road': 0.1158905563486399, 'residential': 0.3170478566896662, 'unlabelled': 0.1600222242049327, 'building': 0.15692133793322227, 'water': 0.019925085256562777}
e3:{'wood': 0.29838557677196365, 'road': 0.10811336240766953, 'residential': 0.303818028052805, 'unlabelled': 0.14888550997956929, 'building': 0.11943965896589669, 'water': 0.02135786382209649}
e4:{'wood': 0.20174671213918546, 'road': 0.1149980525899564, 'residential': 0.35344977856860416, 'unlabelled': 0.1459538335310399, 'building': 0.16415052985369719, 'water': 0.019701093317516807}
e5:{'wood': 0.1725235022950302, 'road': 0.11293436609686605, 'residential': 0.3414093720322888, 'unlabelled': 0.16904109488762292, 'building': 0.16781921098448874, 'water': 0.036272453703703696}
0: ['berl', 'lond', 'mail', 'buka', 'kiew'] 1: ['mosk', 'sama', 'rom_', 'sofi', 'kasa'] 2: ['prag', 'wien', 'madr', 'muni', 'char'] 3: ['pari', 'belg', 'barc', 'hamb', 'ista'] 4: ['buda', 'nisc', 'wars', 'mins', 'st_p']
- [3053, 2954, 3036, 2917, 2941]
- fold splitting with std 54
e1: {'residential': 0.303264278305492, 'wood': 0.18885067420024038, 'building': 0.1835410725333919, 'unlabelled': 0.16669832314299232, 'road': 0.12346273428685832, 'water': 0.03418291753102592}
e2:{'residential': 0.28652251937109763, 'wood': 0.1711001363499586, 'building': 0.17061893665839142, 'unlabelled': 0.2008524637027011, 'road': 0.1376497959452343, 'water': 0.033256147972617174}
e3:{'residential': 0.1989877680793447, 'wood': 0.31977545747328323, 'building': 0.16068834632557463, 'unlabelled': 0.1750524493119603, 'road': 0.12863185203484148, 'water': 0.016864126774996345}
e4:{'residential': 0.2110840827714924, 'wood': 0.25849840589646894, 'building': 0.19598418561688194, 'unlabelled': 0.18470212451910234, 'road': 0.11392435912086225, 'water': 0.03580684207519138}
e5:{'residential': 0.2490577921341947, 'wood': 0.19677026143790854, 'building': 0.16059237504250223, 'unlabelled': 0.1990937436246172, 'road': 0.14519783803694875, 'water': 0.049287989723827856}
- train: 480, test: 60, val: 60
- distrib = {0: 0.615830, 1: 0.038690, 2: 0.004052, 3: 0.008817, 4: 0.029346, 5: 0.108000, 6: 0.189018, 7: 0.005066, 8: 0.001049, 9: 1.176852e-05, 10: 0.000120}
- median frequency balancing:
- 0: 0.014317
- 1: 0.227888
- 3: 2.175962
- 4: 1.000000
- 5: 0.300450
- 6: 0.081639
- 7: 0.046646
- 8: 1.740426
- 9: 8.405148
- 10: 749.202109
- 11: 73.475000}
- train: 264, test: 15, val: 15
-
vaihingen weights
-
{0: 0.27496586356764924, 1: 0.2600085789871505, 2: 0.20469702380952381, 3: 0.23992597316704442, 4: 0.012058994708994705, 5: 0.008343565759637187}
-
median frequency balancing:
-
0: 0.808506
-
1: 0.855016
-
2: 1.086051
-
3: 0.926584
-
4: 18.435326
-
5: 26.644663