Pausing for now, train loop should now work and adding some tests

utils/eval_utils.py

@@ -28,9 +28,13 @@ def evaluate(model, data_loader, device):
             images = list(image.to(device) for image in images)
             targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
 
-            # In eval mode with targets, Mask R-CNN should still return losses
-            # If it returned predictions, logic here would change to process predictions
+            # To handle the different behavior of Mask R-CNN in eval mode,
+            # we explicitly reset the model to training mode to compute losses,
+            # then switch back to eval mode for the rest of the evaluation
+            model.train()
             loss_dict = model(images, targets)
+            model.eval()
+
             losses = sum(loss for loss in loss_dict.values())
             loss_value = losses.item()
             total_loss += loss_value