Update retinaNet + RCNN to use 'classification' (#1313)

Author: LukeWood

keras_cv/models/object_detection/faster_rcnn.py

@@ -429,9 +429,9 @@ def compile(
         self.weight_decay = weight_decay
         losses = {
             "box": self.box_loss,
-            "cls": self.cls_loss,
+            "classification": self.cls_loss,
             "rpn_box": self.rpn_box_loss,
-            "rpn_cls": self.rpn_cls_loss,
+            "rpn_classification": self.rpn_cls_loss,
         }
         super().compile(loss=losses, **kwargs)
 
@@ -466,21 +466,21 @@ def compute_loss(self, images, boxes, classes, training):
         box_pred, cls_pred = self._call_rcnn(rois, feature_map, training=training)
         y_true = {
             "rpn_box": rpn_box_targets,
-            "rpn_cls": rpn_cls_targets,
+            "rpn_classification": rpn_cls_targets,
             "box": box_targets,
-            "cls": cls_targets,
+            "classification": cls_targets,
         }
         y_pred = {
             "rpn_box": rpn_box_pred,
-            "rpn_cls": rpn_cls_pred,
+            "rpn_classification": rpn_cls_pred,
             "box": box_pred,
-            "cls": cls_pred,
+            "classification": cls_pred,
         }
         weights = {
             "rpn_box": rpn_box_weights,
-            "rpn_cls": rpn_cls_weights,
+            "rpn_classification": rpn_cls_weights,
             "box": box_weights,
-            "cls": cls_weights,
+            "classification": cls_weights,
         }
         return super().compute_loss(
             x=images, y=y_true, y_pred=y_pred, sample_weight=weights

keras_cv/models/object_detection/retina_net/retina_net.py

@@ -312,7 +312,7 @@ def compile(
         self.weight_decay = weight_decay
         losses = {
             "box": self.box_loss,
-            "cls": self.classification_loss,
+            "classification": self.classification_loss,
         }
         super().compile(loss=losses, **kwargs)
 
@@ -350,15 +350,15 @@ def compute_loss(self, images, boxes, classes, training):
         box_weights = positive_mask / normalizer
         y_true = {
             "box": boxes,
-            "cls": cls_labels,
+            "classification": cls_labels,
         }
         y_pred = {
             "box": box_pred,
-            "cls": cls_pred,
+            "classification": cls_pred,
         }
         sample_weights = {
             "box": box_weights,
-            "cls": cls_weights,
+            "classification": cls_weights,
         }
         return super().compute_loss(
             x=images, y=y_true, y_pred=y_pred, sample_weight=sample_weights