Fix InceptionV3 and add Codecov

.github/workflows/actions.yml

@@ -45,6 +45,13 @@ jobs:
       - name: Test with pytest
         run: |
           pytest
+          coverage xml -o coverage.xml
+      - name: Upload coverage reports to Codecov
+        uses: codecov/codecov-action@v3
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: coverage.xml
+          flags: kimm,kimm-${{ matrix.backend }}
 
   format:
     name: Check the code format

README.md

@@ -1,5 +1,16 @@
+<!-- markdownlint-disable MD033 -->
+<!-- markdownlint-disable MD041 -->
+
 # Keras Image Models
 
+<div align="center">
+<img width="50%" src="docs/banner/kimm.png" alt="KIMM">
+
+[![PyPI](https://img.shields.io/pypi/v/kimm)](https://pypi.org/project/kimm/)
+[![Contributions Welcome](https://img.shields.io/badge/contributions-welcome-brightgreen.svg?style=flat)](https://github.com/james77777778/kimm/issues)
+[![codecov](https://codecov.io/gh/james77777778/kimm/graph/badge.svg?token=eEha1SR80D)](https://codecov.io/gh/james77777778/kimm)
+</div>
+
 ## Description
 
 **K**eras **Im**age **M**odels (`kimm`) is a collection of image models, blocks and layers written in Keras 3. The goal is to offer SOTA models with pretrained weights in a user-friendly manner.

kimm/models/inception_v3.py

@@ -299,8 +299,14 @@ def __init__(
     ):
         kwargs = self.fix_config(kwargs)
         if weights == "imagenet":
-            has_aux_logits = False
-            file_name = "inceptionv3_inception_v3.gluon_in1k.keras"
+            if has_aux_logits:
+                file_name = (
+                    "inceptionv3_inception_v3.gluon_in1k_aux_logits.keras"
+                )
+            else:
+                file_name = (
+                    "inceptionv3_inception_v3.gluon_in1k_no_aux_logits.keras"
+                )
             kwargs["weights_url"] = f"{self.default_origin}/{file_name}"
         super().__init__(
             has_aux_logits,

pyproject.toml

@@ -33,6 +33,7 @@ tests = [
     "black",
     "pytest",
     "pytest-cov",
+    "coverage",
 ]
 examples = ["opencv-python", "matplotlib"]
 

tools/convert_inception_v3_from_timm.py

@@ -17,30 +17,36 @@
 
 timm_model_names = [
     "inception_v3.gluon_in1k",
+    "inception_v3.gluon_in1k",
 ]
 keras_model_classes = [
     inception_v3.InceptionV3,
+    inception_v3.InceptionV3,
 ]
+has_aux_logits_list = [True, False]
 
-for timm_model_name, keras_model_class in zip(
-    timm_model_names, keras_model_classes
+for timm_model_name, keras_model_class, has_aux_logits in zip(
+    timm_model_names,
+    keras_model_classes,
+    has_aux_logits_list,
 ):
     """
     Prepare timm model and keras model
     """
     input_shape = [299, 299, 3]
     torch_model = timm.create_model(
-        timm_model_name, pretrained=True, aux_logits=False
+        timm_model_name, pretrained=True, aux_logits=has_aux_logits
     )
     torch_model = torch_model.eval()
     trainable_state_dict, non_trainable_state_dict = separate_torch_state_dict(
         torch_model.state_dict()
     )
     keras_model = keras_model_class(
-        has_aux_logits=False,
+        has_aux_logits=has_aux_logits,
         input_shape=input_shape,
         include_preprocessing=False,
         classifier_activation="linear",
+        weights=None,
     )
     trainable_weights, non_trainable_weights = separate_keras_weights(
         keras_model
@@ -129,17 +135,33 @@
     np.random.seed(2023)
     keras_data = np.random.uniform(size=[1] + input_shape).astype("float32")
     torch_data = torch.from_numpy(np.transpose(keras_data, [0, 3, 1, 2]))
-    torch_y = torch_model(torch_data)
-    keras_y = keras_model(keras_data, training=False)
-    torch_y = torch_y.detach().cpu().numpy()
-    keras_y = keras.ops.convert_to_numpy(keras_y)
-    np.testing.assert_allclose(torch_y, keras_y, atol=1e-5)
+    if has_aux_logits:
+        torch_y = torch_model(torch_data)[0]
+        keras_y = keras_model(keras_data, training=False)[0]
+        torch_y = torch_y.detach().cpu().numpy()
+        keras_y = keras.ops.convert_to_numpy(keras_y)
+        np.testing.assert_allclose(torch_y, keras_y, atol=1e-5)
+    else:
+        torch_y = torch_model(torch_data)
+        keras_y = keras_model(keras_data, training=False)
+        torch_y = torch_y.detach().cpu().numpy()
+        keras_y = keras.ops.convert_to_numpy(keras_y)
+        np.testing.assert_allclose(torch_y, keras_y, atol=1e-5)
     print(f"{keras_model_class.__name__}: output matched!")
 
     """
     Save converted model
     """
     os.makedirs("exported", exist_ok=True)
-    export_path = f"exported/{keras_model.name.lower()}_{timm_model_name}.keras"
+    if has_aux_logits:
+        export_path = (
+            f"exported/{keras_model.name.lower()}_{timm_model_name}_"
+            "aux_logits.keras"
+        )
+    else:
+        export_path = (
+            f"exported/{keras_model.name.lower()}_{timm_model_name}_"
+            "no_aux_logits.keras"
+        )
     keras_model.save(export_path)
     print(f"Export to {export_path}")