Fix for relative data augmentation and early stopping frequency >= 1

Calamari-OCR · Feb 8, 2019 · 8a2857b · 8a2857b
1 parent 9528da7
commit 8a2857b
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 1 deletion.
diff --git a/calamari_ocr/ocr/datasets/input_dataset.py b/calamari_ocr/ocr/datasets/input_dataset.py
@@ -58,7 +58,7 @@ def epoch_size(self):
         if self.data_augmentation_amount >= 1:
             return int(len(self) * self.data_augmentation_amount)
 
-        return 1 / (1 - self.data_augmentation_amount)
+        return int(1 / (1 - self.data_augmentation_amount) * len(self))
 
     def preload(self, processes=1, progress_bar=False):
         print("Preloading dataset type {} with size {}".format(self.dataset.mode, len(self)))

diff --git a/calamari_ocr/ocr/trainer.py b/calamari_ocr/ocr/trainer.py
@@ -218,6 +218,7 @@ def _run_train(self, train_net, test_net, codec, train_start_time, progress_bar)
             early_stopping_frequency = int(early_stopping_frequency * iters_per_epoch)  # relative to epochs
         else:
             early_stopping_frequency = int(early_stopping_frequency)
+        early_stopping_frequency = max(1, early_stopping_frequency)
 
         if checkpoint_frequency < 0:
             checkpoint_frequency = early_stopping_frequency