feat: model_converter maskrcnn support (#567)

* maskrcnn

Co-authored-by: Copilot <copilot@github.com>

* small refactor

Co-authored-by: Copilot <copilot@github.com>

* mypy

Co-authored-by: Copilot <copilot@github.com>

* unit tests

Co-authored-by: Copilot <copilot@github.com>

* ruff

Co-authored-by: Copilot <copilot@github.com>

* ruff

* pytest github actions

---------

Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Ivan Zakharov <ivan.zakharov@intel.com>

Author: 3 people

.github/workflows/pre_commit.yml

@@ -152,9 +152,12 @@ jobs:
 
       - *install-dependencies
 
-      - name: Run python unit tests
+      - name: Run model_api unit tests
         run: uv --directory model_api run pytest tests/unit --cov
 
+      - name: Run model_converter unit tests
+        run: uv --directory model_converter run --group tests pytest tests/unit --cov
+
       - &prepare-test-data
         name: Prepare test data
         run: |

model_converter/examples/config.json

@@ -618,6 +618,35 @@
       "scale_values": "58.395 57.12 57.375",
       "license": "apache-2.0",
       "license_link": "https://spdx.org/licenses/Apache-2.0.html"
+    },
+    {
+      "model_short_name": "maskrcnn_resnet50_fpn",
+      "model_class_name": "torchvision.models.detection.maskrcnn_resnet50_fpn",
+      "model_library": "torchvision",
+      "model_full_name": "Mask R-CNN ResNet-50 FPN",
+      "description": "Mask R-CNN with a ResNet-50-FPN backbone trained on COCO for object detection and instance segmentation",
+      "docs": "https://docs.pytorch.org/vision/main/models/generated/torchvision.models.detection.maskrcnn_resnet50_fpn.html",
+      "weights_url": "https://download.pytorch.org/models/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth",
+      "input_shape": [1, 3, 800, 800],
+      "input_names": ["image"],
+      "output_names": ["boxes", "labels", "masks"],
+      "model_params": null,
+      "model_type": "MaskRCNN",
+      "reverse_input_channels": true,
+      "mean_values": "0 0 0",
+      "scale_values": "255 255 255",
+      "resize_type": "fit_to_window_letterbox",
+      "pad_value": 0,
+      "input_dtype": "u8",
+      "confidence_threshold": 0.5,
+      "postprocess_semantic_masks": true,
+      "nms_execute": false,
+      "iou_threshold": 0.5,
+      "agnostic_nms": false,
+      "nms_max_predictions": 200,
+      "license": "bsd-3-clause",
+      "license_link": "https://spdx.org/licenses/BSD-3-Clause.html",
+      "labels": "COCO_V1"
     }
   ]
 }

model_converter/pyproject.toml

@@ -178,6 +178,9 @@ fixable = ["ALL"]
 unfixable = []
 dummy-variable-rgx = "^(_+|(_+[a-zA-Z0-9_]*[a-zA-Z0-9]+?))$"
 
+[tool.ruff.lint.per-file-ignores]
+"**/tests/**/*.py" = ["SLF001", "FBT003"]
+
 [tool.ruff.lint.mccabe]
 max-complexity = 15
 
@@ -193,6 +196,16 @@ notice-rgx = """
 [tool.bandit]
 skips = ["B101", "B310"]
 
+[tool.pytest.ini_options]
+pythonpath = ["src"]
+
+[tool.coverage.run]
+source = ["model_converter"]
+
 [tool.coverage.report]
-fail_under = 45
+fail_under = 100
 show_missing = true
+exclude_lines = [
+    "if __name__ == .__main__.",
+    "if TYPE_CHECKING:",
+]

model_converter/src/model_converter/__init__.py

@@ -5,6 +5,6 @@
 
 """Tools for converting models to OpenVINO IR."""
 
-from .model_converter import ModelConverter, list_models, main
+from .cli import ModelConverter, list_models, main
 
 __all__ = ["ModelConverter", "list_models", "main"]

model_converter/src/model_converter/__main__.py

@@ -5,7 +5,8 @@
 
 """Run the model converter with ``python -m model_converter``."""
 
-from .model_converter import main
+import sys
 
-if __name__ == "__main__":
-    raise SystemExit(main())
+from model_converter.cli import main
+
+sys.exit(main())

model_converter/src/model_converter/adapters/__init__.py

@@ -0,0 +1,42 @@
+#
+# Copyright (C) 2026 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+
+"""Export adapters for different model types."""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from model_converter.adapters.base import ExportAdapter
+from model_converter.adapters.maskrcnn import TorchvisionMaskRCNNExportAdapter
+
+if TYPE_CHECKING:
+    import torch
+
+_ADAPTER_REGISTRY: dict[str, type[ExportAdapter]] = {
+    "maskrcnn": TorchvisionMaskRCNNExportAdapter,
+}
+
+
+def get_adapter(model_type: str, model: torch.nn.Module) -> torch.nn.Module:
+    """
+    Get the appropriate export adapter for a model type.
+
+    If no adapter is registered for the model type, returns the model unchanged.
+
+    Args:
+        model_type: Model type string (e.g., "MaskRCNN")
+        model: The PyTorch model to adapt
+
+    Returns:
+        Adapted model (or original model if no adapter needed)
+    """
+    adapter_class = _ADAPTER_REGISTRY.get(model_type.lower())
+    if adapter_class is not None:
+        return adapter_class(model)
+    return model
+
+
+__all__ = ["ExportAdapter", "TorchvisionMaskRCNNExportAdapter", "get_adapter"]

model_converter/src/model_converter/adapters/base.py

@@ -0,0 +1,16 @@
+#
+# Copyright (C) 2026 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+
+"""Base export adapter interface."""
+
+import torch.nn as nn
+
+
+class ExportAdapter(nn.Module):
+    """Base class for export adapters that reshape model outputs for Model API."""
+
+    def __init__(self, model: nn.Module):
+        super().__init__()
+        self.model = model

model_converter/src/model_converter/adapters/maskrcnn.py

@@ -0,0 +1,31 @@
+#
+# Copyright (C) 2026 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+
+"""Mask R-CNN export adapter for TorchVision models."""
+
+from collections import OrderedDict
+
+import torch
+
+from model_converter.adapters.base import ExportAdapter
+
+
+class TorchvisionMaskRCNNExportAdapter(ExportAdapter):
+    """Adapt TorchVision Mask R-CNN to the Model API MaskRCNN output contract."""
+
+    def forward(self, images: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """Return boxes-with-scores, shifted labels, and raw masks for one image."""
+        image_list = [images[0]]
+        transformed_images, _ = self.model.transform(image_list, None)
+        features = self.model.backbone(transformed_images.tensors)
+        if isinstance(features, torch.Tensor):
+            features = OrderedDict([("0", features)])
+        proposals, _ = self.model.rpn(transformed_images, features, None)
+        predictions, _ = self.model.roi_heads(features, proposals, transformed_images.image_sizes, None)
+        prediction = predictions[0]
+        boxes = torch.cat((prediction["boxes"], prediction["scores"].unsqueeze(1)), dim=1)
+        labels = prediction["labels"] - 1
+        masks = prediction["masks"].squeeze(1)
+        return boxes, labels, masks