Revert "Make input a dictionary for multi-modal object detection (#95)"

examples/carla_overhead_object_detection/configs/experiment/ArmoryCarlaOverObjDet_TorchvisionFasterRCNN.yaml

@@ -2,7 +2,6 @@
 
 defaults:
   - COCO_TorchvisionFasterRCNN
-  - override /model/[email protected]: tuple_tensorizer_normalizer
   - override /datamodule: armory_carla_over_objdet_perturbable_mask
 
 task_name: "ArmoryCarlaOverObjDet_TorchvisionFasterRCNN"

mart/configs/model/torchvision_object_detection.yaml

@@ -1,7 +1,6 @@
 # We simply wrap a torchvision object detection model for validation.
 defaults:
   - modular
-  - /model/[email protected]: tuple_normalizer
 
 training_step_log:
   loss: "loss"
@@ -13,6 +12,13 @@ test_sequence: ???
 output_preds_key: "losses_and_detections.eval"
 
 modules:
+  preprocessor:
+    _target_: mart.transforms.TupleTransforms
+    transforms:
+      _target_: torchvision.transforms.Normalize
+      mean: 0
+      std: 255
+
   losses_and_detections:
     # Return losses in the training mode and predictions in the eval mode in one pass.
     _target_: mart.models.DualMode

mart/datamodules/coco.py

@@ -68,23 +68,6 @@ def _load_target(self, id: int) -> List[Any]:
 
         return {"image_id": id, "file_name": file_name, "annotations": annotations}
 
-    def __getitem__(self, index: int):
-        """Override __getitem__() to dictionarize input for multi-modality datasets.
-
-        This runs after _load_image() and transforms(), while transforms() typically converts
-        images to tensors.
-        """
-
-        image, target_dict = super().__getitem__(index)
-
-        # Convert multi-modal input to a dictionary.
-        if self.modalities is not None:
-            # We assume image is a multi-channel tensor, with each modality including 3 channels.
-            assert image.shape[0] == len(self.modalities) * 3
-            image = dict(zip(self.modalities, image.split(3)))
-
-        return image, target_dict
-
 
 # Source: https://github.com/pytorch/vision/blob/dc07ac2add8285e16a716564867d0b4b953f6735/references/detection/utils.py#L203
 def collate_fn(batch):

mart/transforms/transforms.py

@@ -7,16 +7,7 @@
 import torch
 from torchvision.transforms import transforms as T
 
-__all__ = [
-    "Denormalize",
-    "Cat",
-    "Permute",
-    "Unsqueeze",
-    "Squeeze",
-    "Chunk",
-    "TupleTransforms",
-    "GetItems",
-]
+__all__ = ["Denormalize", "Cat", "Permute", "Unsqueeze", "Squeeze", "Chunk", "TupleTransforms"]
 
 
 class Denormalize(T.Normalize):
@@ -90,14 +81,3 @@ def __init__(self, transforms):
     def forward(self, x_tuple):
         output_tuple = tuple(self.transforms(x) for x in x_tuple)
         return output_tuple
-
-
-class GetItems:
-    """Get a list of values with a list of keys from a dictionary."""
-
-    def __init__(self, keys):
-        self.keys = keys
-
-    def __call__(self, x):
-        x_list = [x[key] for key in self.keys]
-        return x_list