feat: pass missing model_info (#472)

* feat: pass missing model_info

* Update README.md

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Fix code-style

* Update tests/accuracy/test_accuracy.py

Co-authored-by: Mariusz Gumowski <mariusz.gumowski@intel.com>

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Mariusz Gumowski <mariusz.gumowski@intel.com>

Author: 3 people

README.md

@@ -59,4 +59,32 @@ model = DetectionModel.create_model("~/.cache/omz/public/ssdlite_mobilenet_v2/FP
 model.save("serialized.xml")
 ```
 
+## Usage with generic OpenVINO models
+
+ModelAPI uses custom field in `rt_info/model_info` section of OpenVINO IR to store metadata required for preprocessing and postprocessing. If you have a generic OpenVINO model without such metadata, you can provide that metadata in `configuration` argument of `create_model()` method:
+
+```python
+from model_api.models import Model
+
+# Create a model wrapper from a compatible model generated by OpenVINO Training Extensions
+model = Model.create_model(
+    "model.xml",
+    configuration={
+        "model_type": "Segmentation",
+        "blur_strength": 1,
+        "labels": ["object"],
+        "soft_threshold": 0.5,
+    }
+)
+
+# Run synchronous inference locally
+result = model(image)  # image is numpy.ndarray
+
+# Print results in model-specific format
+print(f"Inference result: {result}")
+
+# Save the model with metadata already embedded (passing configuration is not required anymore)
+model.save("serialized_with_metadata.xml")
+```
+
 For more details please refer to the [examples](https://github.com/openvinotoolkit/model_api/tree/master/examples) of this project.

src/model_api/models/model.py

@@ -234,7 +234,7 @@ def create_model(
                     ["model_info", "model_type"],
                 ).astype(str)
             except RuntimeError:
-                model_type = cls.detect_model_type(inference_adapter)
+                model_type = configuration.get("model_type", cls.detect_model_type(inference_adapter))
         Model = cls.get_model_class(model_type)
         return Model(inference_adapter, configuration, preload)
 

tests/accuracy/public_scope.json

@@ -35,6 +35,24 @@
       }
     ]
   },
+  {
+    "name": "otx_models/Lite-hrnet-s_mod2-without-model-info.xml",
+    "type": "SegmentationModel",
+    "configuration": {
+      "blur_strength": 1,
+      "labels": ["object"],
+      "model_type": "Segmentation",
+      "soft_threshold": 0.5
+    },
+    "test_data": [
+      {
+        "image": "coco128/images/train2017/000000000074.jpg",
+        "reference": [
+          "0: 0.563, 1: 0.437, [426,640,2], [0], [0]; object: 0.520, 26, 0, object: 0.530, 42, 0, object: 0.501, 4, 0, object: 0.507, 27, 0, object: 0.503, 8, 0, object: 0.502, 6, 0, object: 0.505, 18, 0, object: 0.504, 13, 0, object: 0.524, 87, 0, object: 0.521, 89, 0, object: 0.757, 2706, 2, "
+        ]
+      }
+    ]
+  },
   {
     "name": "otx_models/Lite-hrnet-s_mod2.onnx",
     "type": "SegmentationModel",

tests/accuracy/test_accuracy.py

@@ -78,7 +78,7 @@ def read_config(fname):
         return json.load(f)
 
 
-def create_models(model_type, model_path, download_dir, force_onnx_adapter=False, device="CPU"):
+def create_models(model_type, model_path, download_dir, force_onnx_adapter=False, device="CPU", configuration=None):
     if model_path.endswith(".onnx") and force_onnx_adapter:
         wrapper_type = model_type.get_model_class(
             load_parameters_from_onnx(onnx.load(model_path))["model_info"]["model_type"],
@@ -92,16 +92,18 @@ def create_models(model_type, model_path, download_dir, force_onnx_adapter=False
         model.load()
         return [model]
 
+    configuration = configuration or {}
+
     models = [
-        model_type.create_model(model_path, device=device, download_dir=download_dir),
+        model_type.create_model(model_path, device=device, download_dir=download_dir, configuration=configuration),
     ]
     if model_path.endswith(".xml"):
         model = create_core().read_model(model_path)
         if model.has_rt_info(["model_info", "model_type"]):
             wrapper_type = model_type.get_model_class(
                 model.get_rt_info(["model_info", "model_type"]).astype(str),
             )
-            model = wrapper_type(OpenvinoAdapter(create_core(), model_path, device=device))
+            model = wrapper_type(OpenvinoAdapter(create_core(), model_path, device=device), configuration=configuration)
             model.load()
             models.append(model)
     return models
@@ -287,6 +289,7 @@ def test_image_models(data, device, dump, result, model_data, results_dir):  # n
         data,
         model_data.get("force_ort", False),
         device=device,
+        configuration=model_data.get("configuration", None),
     ):
         if "tiler" in model_data:
             if "extra_model" in model_data: