[saliencymap_evaluator] Improve TODO comments with more ideas

src/ptbench/engine/saliencymap_evaluator.py

@@ -93,8 +93,6 @@ def _extract_bounding_box(
         * width (pixels)
         * height (pixels)
     """
-    # opencv implementation:
-    # x, y, w, h = cv2.boundingRect(mask.astype(numpy.uint8))
     x, y, x2, y2 = torchvision.ops.masks_to_boxes(torch.tensor(mask)[None, :])[
         0
     ]
@@ -330,14 +328,15 @@ def run(
 
     retval: dict[str, list[typing.Any]] = {}
 
+    # TODO: This loads the images from the dataset, but they are not useful at
+    # this point.  Possibly using the contents of ``datamodule.splits`` can
+    # substantially speed this up.
     for dataset_name, dataset_loader in datamodule.predict_dataloader().items():
         logger.info(
             f"Estimating interpretability metrics for dataset `{dataset_name}`..."
         )
         retval[dataset_name] = []
 
-        # TODO: This loads the images from the dataset, but they are not useful at
-        # this point...
         for sample in tqdm(
             dataset_loader, desc="batches", leave=False, disable=None
         ):