Added application of inverse transforms and removal of dummy predictions

Signed-off-by: Jim Martens <github@2martens.de>

src/twomartens/masterthesis/ssd.py

@@ -35,6 +35,7 @@ Functions:
     train(...): trains the SSD/DropoutSSD on a given data set
 """
 import os
+import pickle
 import time
 from typing import Dict
 from typing import Optional
@@ -191,28 +192,53 @@ def _predict_one_epoch(dataset: tf.data.Dataset,
         if use_dropout:
             for _ in range(forward_passes_per_image):
                 result = np.array(ssd(inputs))
+                result_filtered = []
+                # iterate over result of images
+                for i in range(result.shape[0]):
+                    # apply inverse transformations to predicted bounding box coordinates
+                    # filter out dummy all-zero results
+                    x_reverse = labels[i, 0, 5]
+                    y_reverse = labels[i, 0, 6]
+                    filtered = result[i][result[i, :, 0] != 0]
+                    filtered[:, 2] *= x_reverse
+                    filtered[:, 4] *= x_reverse
+                    filtered[:, 3] *= y_reverse
+                    filtered[:, 5] *= y_reverse
+                    result_filtered.append(filtered)
+                result = result_filtered
                 decoded_predictions_batch.append(result)
                 del result
         else:
             result = np.array(ssd(inputs))
+            result_filtered = []
+            # iterate over result of images
+            for i in range(result.shape[0]):
+                # apply inverse transformations to predicted bounding box coordinates
+                # filter out dummy all-zero results
+                x_reverse = labels[i, 0, 5]
+                y_reverse = labels[i, 0, 6]
+                filtered = result[i][result[i, :, 0] != 0]
+                filtered[:, 2] *= x_reverse
+                filtered[:, 4] *= x_reverse
+                filtered[:, 3] *= y_reverse
+                filtered[:, 5] *= y_reverse
+                result_filtered.append(filtered)
+            result = result_filtered
             decoded_predictions_batch.append(result)
             del result
-
+        
         # save predictions batch-wise to prevent memory problems
         if nr_digits is not None:
             counter_str = str(counter).zfill(nr_digits)
-            filename = f"{output_file}-{counter_str}.npy"
-            label_filename = f"{label_output_file}-{counter_str}.npy"
+            filename = f"{output_file}-{counter_str}.bin"
+            label_filename = f"{label_output_file}-{counter_str}.bin"
         else:
-            filename = f"{output_file}-{counter:d}.npy"
-            label_filename = f"{label_output_file}-{counter:d}.npy"
+            filename = f"{output_file}-{counter:d}.bin"
+            label_filename = f"{label_output_file}-{counter:d}.bin"
         
         with open(filename, 'wb') as file, open(label_filename, 'wb') as label_file:
-            decoded_predictions_batch_np = np.array(decoded_predictions_batch)
-            del decoded_predictions_batch
-            np.save(file, decoded_predictions_batch_np, allow_pickle=False, fix_imports=False)
-            del decoded_predictions_batch_np
-            np.save(label_file, labels, allow_pickle=False, fix_imports=False)
+            pickle.dump(decoded_predictions_batch, file)
+            pickle.dump(labels, label_file)
         
         counter += 1