Update

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

__init__.py

@@ -12,10 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import os
+# import os
 
-# For requirements.
-try:
-    import timm
-except ModuleNotFoundError:
-    os.system('pip install timm')
+# # For requirements.
+# try:
+#     import timm
+# except ModuleNotFoundError:
+#     os.system('pip install timm')

pytorch/__init__.py

@@ -11,3 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+import os
+
+# For requirements.
+try:
+    import timm
+except ModuleNotFoundError:
+    os.system('pip install timm')

vit_image_embedding.py

@@ -20,12 +20,12 @@ from PIL import Image
 import torch
 from timm.data import resolve_data_config
 from timm.data.transforms_factory import create_transform
-
+import numpy
 
 from towhee.operator import Operator
 
 
-class VisionTransformerEmbeddingOperator(Operator):
+class VitImageEmbedding(Operator):
     """
     Embedding extractor using ViT.
     Args:
@@ -40,13 +40,13 @@ class VisionTransformerEmbeddingOperator(Operator):
         super().__init__()
         sys.path.append(str(Path(__file__).parent))
         if framework == 'pytorch':
-            from vit_embedding.pytorch.model import Model
+            from pytorch.model import Model
         self.model = Model(model_name, weights_path)
         config = resolve_data_config({}, model=self.model._model)
         self.tfms = create_transform(**config)
 
-    def __call__(self, img_path: str) -> NamedTuple('Outputs', [('embedding', torch.Tensor)]):
-        Outputs = NamedTuple('Outputs', [('embedding', torch.Tensor)])
+    def __call__(self, img_path: str) -> NamedTuple('Outputs', [('feature_vector', numpy.ndarray)]):
+        Outputs = NamedTuple('Outputs', [('feature_vector', numpy.ndarray)])
         img = self.tfms(Image.open(img_path)).unsqueeze(0)
         features = self.model(img)
         return Outputs(features.flatten().detach().numpy())