Assorted improvements

armory/baseline_models/tf_graph/audio_resnet50.py

@@ -53,6 +53,7 @@ def get_art_model(
 
     loss_object = losses.SparseCategoricalCrossentropy()
 
+    @tf.function
     def train_step(model, samples, labels):
         with tf.GradientTape() as tape:
             predictions = model(samples, training=True)

armory/data/adversarial_datasets.py

@@ -661,15 +661,15 @@ def carla_over_obj_det_dev(
     **kwargs,
 ):
     """
-    Dev set for CARLA object detection dataset, containing RGB and depth channels. The dev
+    Dev set for CARLA overhead object detection dataset, containing RGB and depth channels. The dev
     set also contains green screens for adversarial patch insertion.
     """
     if "class_ids" in kwargs:
         raise ValueError(
-            "Filtering by class is not supported for the carla_obj_det_dev dataset"
+            "Filtering by class is not supported for the carla_over_obj_det_dev dataset"
         )
     if batch_size != 1:
-        raise ValueError("carla_obj_det_dev batch size must be set to 1")
+        raise ValueError("carla_over_obj_det_dev batch size must be set to 1")
 
     modality = kwargs.pop("modality", "rgb")
     if modality not in ["rgb", "depth", "both"]:
@@ -729,15 +729,15 @@ def carla_over_obj_det_test(
     **kwargs,
 ):
     """
-    Dev set for CARLA object detection dataset, containing RGB and depth channels. The test
+    Test set for CARLA overhead object detection dataset, containing RGB and depth channels. The test
     set also contains green screens for adversarial patch insertion.
     """
     if "class_ids" in kwargs:
         raise ValueError(
-            "Filtering by class is not supported for the carla_obj_det_test dataset"
+            "Filtering by class is not supported for the carla_over_obj_det_test dataset"
         )
     if batch_size != 1:
-        raise ValueError("carla_obj_det_test batch size must be set to 1")
+        raise ValueError("carla_over_obj_det_test batch size must be set to 1")
 
     modality = kwargs.pop("modality", "rgb")
     if modality not in ["rgb", "depth", "both"]:
@@ -924,7 +924,7 @@ def carla_video_tracking_dev(
             "Filtering by class is not supported for the carla_video_tracking_dev dataset"
         )
     if batch_size != 1:
-        raise ValueError("carla_obj_det_dev batch size must be set to 1")
+        raise ValueError("carla_video_tracking_dev batch size must be set to 1")
 
     if max_frames:
         clip = datasets.ClipFrames(max_frames)
@@ -975,10 +975,10 @@ def carla_video_tracking_test(
     """
     if "class_ids" in kwargs:
         raise ValueError(
-            "Filtering by class is not supported for the carla_video_tracking_dev dataset"
+            "Filtering by class is not supported for the carla_video_tracking_test dataset"
         )
     if batch_size != 1:
-        raise ValueError("carla_obj_det_dev batch size must be set to 1")
+        raise ValueError("carla_video_tracking_test batch size must be set to 1")
 
     if max_frames:
         clip = datasets.ClipFrames(max_frames)

armory/data/datasets.py

@@ -1034,11 +1034,11 @@ def carla_over_obj_det_train(
     **kwargs,
 ) -> ArmoryDataGenerator:
     """
-    Training set for CARLA object detection dataset, containing RGB and depth channels.
+    Training set for CARLA overhead object detection dataset, containing RGB and depth channels.
     """
     if "class_ids" in kwargs:
         raise ValueError(
-            "Filtering by class is not supported for the carla_obj_det_train dataset"
+            "Filtering by class is not supported for the carla_over_obj_det_train dataset"
         )
     modality = kwargs.pop("modality", "rgb")
     if modality not in ["rgb", "depth", "both"]:

armory/utils/config_loading.py

@@ -79,6 +79,11 @@ def load_dataset(dataset_config, *args, num_batches=None, check_run=False, **kwa
     if check_run:
         return EvalGenerator(dataset, num_eval_batches=1)
     if num_batches:
+        if num_batches > dataset.batches_per_epoch:
+            # since num-eval-batches only applies at test time, we can assume there is only 1 epoch
+            raise ValueError(
+                f"{num_batches} eval batches were requested, but dataset has only {dataset.batches_per_epoch} batches of size {dataset.batch_size}"
+            )
         return EvalGenerator(dataset, num_eval_batches=num_batches)
     return dataset