Allowing for input resolutions other than 224x224 in CenterNet ResNet models.

PiperOrigin-RevId: 327521336

research/object_detection/models/center_net_resnet_feature_extractor.py

@@ -46,10 +46,12 @@ class CenterNetResnetFeatureExtractor(CenterNetFeatureExtractor):
         channel_means=channel_means, channel_stds=channel_stds,
         bgr_ordering=bgr_ordering)
     if resnet_type == 'resnet_v2_101':
-      self._base_model = tf.keras.applications.ResNet101V2(weights=None)
+      self._base_model = tf.keras.applications.ResNet101V2(weights=None,
+                                                           include_top=False)
       output_layer = 'conv5_block3_out'
     elif resnet_type == 'resnet_v2_50':
-      self._base_model = tf.keras.applications.ResNet50V2(weights=None)
+      self._base_model = tf.keras.applications.ResNet50V2(weights=None,
+                                                          include_top=False)
       output_layer = 'conv5_block3_out'
     else:
       raise ValueError('Unknown Resnet Model {}'.format(resnet_type))

research/object_detection/models/center_net_resnet_feature_extractor_tf2_test.py

@@ -31,11 +31,11 @@ class CenterNetResnetFeatureExtractorTest(test_case.TestCase):
     model = center_net_resnet_feature_extractor.\
                 CenterNetResnetFeatureExtractor('resnet_v2_101')
     def graph_fn():
-      img = np.zeros((8, 224, 224, 3), dtype=np.float32)
+      img = np.zeros((8, 512, 512, 3), dtype=np.float32)
       processed_img = model.preprocess(img)
       return model(processed_img)
     outputs = self.execute(graph_fn, [])
-    self.assertEqual(outputs.shape, (8, 56, 56, 64))
+    self.assertEqual(outputs.shape, (8, 128, 128, 64))
 
   def test_output_size_resnet50(self):
     """Verify that shape of features returned by the backbone is correct."""

research/object_detection/models/center_net_resnet_v1_fpn_feature_extractor.py

@@ -71,13 +71,15 @@ class CenterNetResnetV1FpnFeatureExtractor(CenterNetFeatureExtractor):
         channel_means=channel_means, channel_stds=channel_stds,
         bgr_ordering=bgr_ordering)
     if resnet_type == 'resnet_v1_50':
-      self._base_model = tf.keras.applications.ResNet50(weights=None)
+      self._base_model = tf.keras.applications.ResNet50(weights=None,
+                                                        include_top=False)
     elif resnet_type == 'resnet_v1_101':
-      self._base_model = tf.keras.applications.ResNet101(weights=None)
+      self._base_model = tf.keras.applications.ResNet101(weights=None,
+                                                         include_top=False)
     elif resnet_type == 'resnet_v1_18':
-      self._base_model = resnet_v1.resnet_v1_18(weights=None)
+      self._base_model = resnet_v1.resnet_v1_18(weights=None, include_top=False)
     elif resnet_type == 'resnet_v1_34':
-      self._base_model = resnet_v1.resnet_v1_34(weights=None)
+      self._base_model = resnet_v1.resnet_v1_34(weights=None, include_top=False)
     else:
       raise ValueError('Unknown Resnet Model {}'.format(resnet_type))
     output_layers = _RESNET_MODEL_OUTPUT_LAYERS[resnet_type]

research/object_detection/models/center_net_resnet_v1_fpn_feature_extractor_tf2_test.py

@@ -40,11 +40,11 @@ class CenterNetResnetV1FpnFeatureExtractorTest(test_case.TestCase,
     model = center_net_resnet_v1_fpn_feature_extractor.\
                 CenterNetResnetV1FpnFeatureExtractor(resnet_type)
     def graph_fn():
-      img = np.zeros((8, 224, 224, 3), dtype=np.float32)
+      img = np.zeros((8, 512, 512, 3), dtype=np.float32)
       processed_img = model.preprocess(img)
       return model(processed_img)
 
-    self.assertEqual(self.execute(graph_fn, []).shape, (8, 56, 56, 64))
+    self.assertEqual(self.execute(graph_fn, []).shape, (8, 128, 128, 64))
 
 
 if __name__ == '__main__':