Use numpy as the inputs.

Author: james77777778

keras_hub/src/models/depth_anything/depth_anything_backbone_test.py

@@ -1,5 +1,5 @@
+import numpy as np
 import pytest
-from keras import ops
 
 from keras_hub.src.models.depth_anything.depth_anything_backbone import (
     DepthAnythingBackbone,
@@ -31,7 +31,7 @@ def setUp(self):
             "head_in_index": -1,
             "feature_keys": ["Stage1", "Stage2", "Stage3", "Stage4"],
         }
-        self.input_data = ops.ones((2, 70, 70, 3))
+        self.input_data = np.ones((2, 70, 70, 3), dtype="float32")
 
     def test_backbone_basics(self):
         self.run_backbone_test(

keras_hub/src/models/depth_anything/depth_anything_depth_estimator_test.py

@@ -1,5 +1,5 @@
+import numpy as np
 import pytest
-from keras import ops
 
 from keras_hub.src.models.depth_anything.depth_anything_backbone import (
     DepthAnythingBackbone,
@@ -30,8 +30,8 @@ def setUp(self):
             image_shape=(70, 70, 3),
             apply_layernorm=True,
         )
-        self.images = ops.ones((2, 70, 70, 3))
-        self.depths = ops.zeros((2, 70, 70, 1))
+        self.images = np.ones((2, 70, 70, 3), dtype="float32")
+        self.depths = np.zeros((2, 70, 70, 1), dtype="float32")
         self.image_converter = DepthAnythingImageConverter(image_size=(70, 70))
         self.preprocessor = DepthAnythingDepthEstimatorPreprocessor(
             self.image_converter
@@ -82,7 +82,7 @@ def test_saved_model(self):
 
     @pytest.mark.extra_large
     def test_all_presets(self):
-        images = ops.ones((2, 518, 518, 3))
+        images = np.ones((2, 518, 518, 3), dtype="float32")
         for preset in DepthAnythingDepthEstimator.presets:
             self.run_preset_test(
                 cls=DepthAnythingDepthEstimator,

keras_hub/src/models/dinov2/dinov2_layers.py

@@ -1,3 +1,4 @@
+import keras
 from keras import backend
 from keras import config
 from keras import initializers
@@ -276,20 +277,24 @@ def get_config(self):
         )
         return config
 
-    def compute_output_shape(self, input_shape):
-        output_shape = [input_shape[0], None, self.hidden_dim]
+    def compute_output_shape(self, inputs_shape):
+        output_shape = [inputs_shape[0], None, self.hidden_dim]
         if self.data_format == "channels_last":
-            if input_shape[1] is not None and input_shape[2] is not None:
-                patch_num = input_shape[1] // self.patch_size
+            if inputs_shape[1] is not None and inputs_shape[2] is not None:
+                patch_num = inputs_shape[1] // self.patch_size
                 # 1 is for cls token.
                 output_shape[1] = 1 + self.num_register_tokens + patch_num**2
         else:
-            if input_shape[2] is not None and input_shape[3] is not None:
-                patch_num = input_shape[2] // self.patch_size
+            if inputs_shape[2] is not None and inputs_shape[3] is not None:
+                patch_num = inputs_shape[2] // self.patch_size
                 # 1 is for cls token.
                 output_shape[1] = 1 + self.num_register_tokens + patch_num**2
         return output_shape
 
+    def compute_output_spec(self, inputs):
+        output_shape = self.compute_output_shape(inputs.shape)
+        return keras.KerasTensor(output_shape, dtype=self.compute_dtype)
+
     @staticmethod
     def _interpolate_position_embeddings(
         position_embeddings,