Update model

Signed-off-by: shiyu22 <shiyu.chen@zilliz.com>

pytorch/resnet50.py → pytorch/model.py

@@ -13,28 +13,28 @@
 # limitations under the License.
 
 
+from typing import NamedTuple
+
+import numpy
 import torch
 import torchvision
 
 
-class Resnet50():
+class Model():
     """
-    PyTorch model for image embedding.
+    PyTorch model class
     """
-    def __init__(self, model_name: str):
-        self.model_name = model_name
+    def __init__(self, model_name):
+        super().__init__()
+        model_func = getattr(torchvision.models, model_name)
+        self._model = model_func(pretrained=True)
+        self._model.eval()
 
-    def load_model(self):
-        """
-        For loading model
-        """
-        model_func = getattr(torchvision.models, self.model_name)
-        self.model = model_func(pretrained=True)
-        self.model.eval()
-        return self.model
-    
-    def train_model(self):
+    def __call__(self, img_tensor: torch.Tensor):
+        return self._model(img_tensor).detach().numpy()
+
+    def train(self):
         """
         For training model
         """
-        pass
+        pass

resnet50_image_embedding.py

@@ -26,13 +26,16 @@ class Resnet50ImageEmbedding(Operator):
     """
     PyTorch model for image embedding.
     """
-    def __init__(self, model_name: str) -> None:
+    def __init__(self, model_name: str, framework: str = 'pytorch') -> None:
         super().__init__()
         sys.path.append(str(Path(__file__).parent))
-        from pytorch.resnet50 import Resnet50
-        resnet50_image_embedding = Resnet50(model_name)
-        self._model = resnet50_image_embedding.load_model()
+        if framework == 'pytorch':
+            from pytorch.model import Model
+        if framework == 'tensorflow':
+            from tensorflow.model import Model
+        self.model = Model(model_name)
 
     def __call__(self, img_tensor: torch.Tensor) -> NamedTuple('Outputs', [('cnn', numpy.ndarray)]):
+        embedding = self.model(img_tensor)
         Outputs = NamedTuple('Outputs', [('cnn', numpy.ndarray)])
-        return Outputs(self._model(img_tensor).detach().numpy())
+        return Outputs(embedding)