Move mpa.deploy to otx.algorithms.common

docs/source/guide/reference/mpa/index.rst

@@ -8,5 +8,4 @@ Model Preparation Algorithm
    classification
    detection
    segmentation
-   deploy
    utils

otx/algorithms/common/adapters/mmdeploy/__init__.py

@@ -0,0 +1,10 @@
+"""Adapters for mmdeploy."""
+# Copyright (C) 2023 Intel Corporation
+#
+# SPDX-License-Identifier: MIT
+
+from .utils.mmdeploy import is_mmdeploy_enabled
+
+__all__ = [
+    "is_mmdeploy_enabled",
+]

otx/mpa/deploy/apis.py → otx/algorithms/common/adapters/mmdeploy/apis.py

@@ -1,3 +1,4 @@
+"""API of otx.algorithms.common.adapters.mmdeploy."""
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
@@ -12,20 +13,23 @@
 
 import mmcv
 import numpy as np
-import onnx
 import torch
 from mmcv.parallel import collate, scatter
 
-from .utils import numpy_2_list
 from .utils.mmdeploy import (
     is_mmdeploy_enabled,
     mmdeploy_init_model_helper,
     update_deploy_cfg,
 )
 from .utils.onnx import prepare_onnx_for_openvino
+from .utils.utils import numpy_2_list
+
+# pylint: disable=too-many-locals
 
 
 class NaiveExporter:
+    """NaiveExporter for non-mmdeploy export."""
+
     @staticmethod
     def export2openvino(
         output_dir: str,
@@ -38,13 +42,15 @@ def export2openvino(
         input_names: Optional[List[str]] = None,
         output_names: Optional[List[str]] = None,
         opset_version: int = 11,
-        dynamic_axes: Dict[Any, Any] = {},
+        dynamic_axes: Optional[Dict[Any, Any]] = None,
         mo_transforms: str = "",
     ):
+        """Function for exporting to openvino."""
         input_data = scatter(collate([input_data], samples_per_gpu=1), [-1])[0]
 
         model = model_builder(cfg)
         model = model.cpu().eval()
+        dynamic_axes = dynamic_axes if dynamic_axes else dict()
 
         onnx_path = NaiveExporter.torch2onnx(
             output_dir,
@@ -108,17 +114,19 @@ def torch2onnx(
         input_names: Optional[List[str]] = None,
         output_names: Optional[List[str]] = None,
         opset_version: int = 11,
-        dynamic_axes: Dict[Any, Any] = {},
+        dynamic_axes: Optional[Dict[Any, Any]] = None,
         verbose: bool = False,
         **onnx_options,
     ) -> str:
+        """Function for torch to onnx exporting."""
 
         img_metas = input_data.get("img_metas")
         numpy_2_list(img_metas)
         imgs = input_data.get("img")
         model.forward = partial(model.forward, img_metas=img_metas, return_loss=False)
 
         onnx_file_name = model_name + ".onnx"
+        dynamic_axes = dynamic_axes if dynamic_axes else dict()
         torch.onnx.export(
             model,
             imgs,
@@ -143,6 +151,7 @@ def onnx2openvino(
         model_name: str = "model",
         **openvino_options,
     ) -> Tuple[str, str]:
+        """Function for onnx to openvino exporting."""
         from otx.mpa.utils import mo_wrapper
 
         mo_args = {
@@ -163,17 +172,15 @@ def onnx2openvino(
 
 if is_mmdeploy_enabled():
     import mmdeploy.apis.openvino as openvino_api
-    from mmdeploy.apis import (
-        build_task_processor,
-        extract_model,
-        get_predefined_partition_cfg,
-        torch2onnx,
-    )
+    from mmdeploy.apis import build_task_processor, extract_model, torch2onnx
     from mmdeploy.apis.openvino import get_input_info_from_cfg, get_mo_options_from_cfg
-    from mmdeploy.core import FUNCTION_REWRITER
-    from mmdeploy.utils import get_backend_config, get_ir_config, get_partition_config
+
+    # from mmdeploy.core import FUNCTION_REWRITER
+    from mmdeploy.utils import get_ir_config, get_partition_config
 
     class MMdeployExporter:
+        """MMdeployExporter for mmdeploy exporting."""
+
         @staticmethod
         def export2openvino(
             output_dir: str,
@@ -183,6 +190,7 @@ def export2openvino(
             *,
             model_name: str = "model",
         ):
+            """Function for exporting to openvino."""
 
             task_processor = build_task_processor(cfg, deploy_cfg, "cpu")
 
@@ -248,6 +256,7 @@ def torch2onnx(
             *,
             model_name: str = "model",
         ) -> str:
+            """Function for torch to onnx exporting."""
             onnx_file_name = model_name + ".onnx"
             torch2onnx(
                 input_data,
@@ -266,6 +275,7 @@ def partition_onnx(
             onnx_path: str,
             partition_cfgs: Union[mmcv.ConfigDict, List[mmcv.ConfigDict]],
         ) -> Tuple[str, ...]:
+            """Function for parition onnx."""
             partitioned_paths = []
 
             if not isinstance(partition_cfgs, list):
@@ -290,6 +300,7 @@ def onnx2openvino(
             *,
             model_name: Optional[str] = None,
         ) -> Tuple[str, str]:
+            """Function for onnx to openvino exporting."""
 
             input_info = get_input_info_from_cfg(deploy_cfg)
             output_names = get_ir_config(deploy_cfg).output_names

otx/mpa/deploy/utils/__init__.py → otx/algorithms/common/adapters/mmdeploy/utils/__init__.py

@@ -1,3 +1,4 @@
+"""Init file for otx.algorithms.common.adapters.mmdeploy.utils."""
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #

otx/mpa/deploy/utils/mmdeploy.py → otx/algorithms/common/adapters/mmdeploy/utils/mmdeploy.py

@@ -1,18 +1,29 @@
+"""Functions for mmdeploy adapters."""
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
 
 import importlib
 
 import onnx
-from mmcv.utils import ConfigDict
 
 
 def is_mmdeploy_enabled():
+    """Checks if the 'mmdeploy' Python module is installed and available for use.
+
+    Returns:
+        bool: True if 'mmdeploy' is installed, False otherwise.
+
+    Example:
+        >>> is_mmdeploy_enabled()
+        True
+    """
     return importlib.util.find_spec("mmdeploy") is not None
 
 
 def mmdeploy_init_model_helper(ctx, model_checkpoint=None, cfg_options=None, **kwargs):
+    """Helper function for initializing a model for inference using the 'mmdeploy' library."""
+
     model_builder = kwargs.pop("model_builder")
     model = model_builder(
         ctx.model_cfg,
@@ -31,12 +42,14 @@ def mmdeploy_init_model_helper(ctx, model_checkpoint=None, cfg_options=None, **k
     return model
 
 
-def update_deploy_cfg(onnx_path, deploy_cfg, mo_options={}):
+def update_deploy_cfg(onnx_path, deploy_cfg, mo_options=None):
+    """Update the 'deploy_cfg' configuration file based on the ONNX model specified by 'onnx_path'."""
+
     from mmdeploy.utils import get_backend_config, get_ir_config
 
     onnx_model = onnx.load(onnx_path)
     ir_config = get_ir_config(deploy_cfg)
-    backend_config = get_backend_config(deploy_cfg)
+    get_backend_config(deploy_cfg)
 
     # update input
     input_names = [i.name for i in onnx_model.graph.input]
@@ -47,6 +60,7 @@ def update_deploy_cfg(onnx_path, deploy_cfg, mo_options={}):
     ir_config["output_names"] = output_names
 
     # update mo options
+    mo_options = mo_options if mo_options else dict()
     deploy_cfg.merge_from_dict({"backend_config": {"mo_options": mo_options}})
 
 

otx/mpa/deploy/utils/onnx.py → otx/algorithms/common/adapters/mmdeploy/utils/onnx.py

@@ -1,3 +1,4 @@
+"""Functions for onnx adapters."""
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
@@ -6,6 +7,7 @@
 
 
 def remove_nodes_by_op_type(onnx_model, op_type):
+    """Remove all nodes of a specified op type from the ONNX model."""
     # TODO: support more nodes
 
     supported_op_types = ["Mark", "Conv", "Gemm"]
@@ -42,6 +44,7 @@ def remove_nodes_by_op_type(onnx_model, op_type):
 
 
 def prepare_onnx_for_openvino(in_path, out_path):
+    """Modify the specified ONNX model to be compatible with OpenVINO by removing 'Mark' op nodes."""
     onnx_model = onnx.load(in_path)
     onnx_model = remove_nodes_by_op_type(onnx_model, "Mark")
     onnx.checker.check_model(onnx_model)

otx/mpa/deploy/utils/operations_domain.py → otx/algorithms/common/adapters/mmdeploy/utils/operations_domain.py

@@ -1,3 +1,4 @@
+"""Add domain function."""
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
@@ -6,4 +7,5 @@
 
 
 def add_domain(name_operator: str) -> str:
+    """Function for adding to DOMAIN_CUSTOM_OPS_NAME."""
     return DOMAIN_CUSTOM_OPS_NAME + "::" + name_operator

otx/mpa/deploy/utils/utils.py → otx/algorithms/common/adapters/mmdeploy/utils/utils.py

@@ -1,3 +1,4 @@
+"""Util functions of otx.algorithms.common.adapters.mmdeploy."""
 # Copyright (C) 2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
@@ -9,6 +10,7 @@
 
 
 def sync_batchnorm_2_batchnorm(module, dim=2):
+    """Syncs the BatchNorm layers in a model to use regular BatchNorm layers."""
     if dim == 1:
         bn = torch.nn.BatchNorm1d
     elif dim == 2:
@@ -48,6 +50,7 @@ def sync_batchnorm_2_batchnorm(module, dim=2):
 
 
 def numpy_2_list(data):
+    """Converts NumPy arrays to Python lists."""
 
     if isinstance(data, np.ndarray):
         return data.tolist()

otx/algorithms/detection/adapters/mmdet/models/detectors/custom_atss_detector.py

@@ -9,10 +9,10 @@
 from mmdet.models.builder import DETECTORS
 from mmdet.models.detectors.atss import ATSS
 
+from otx.algorithms.common.adapters.mmdeploy.utils import is_mmdeploy_enabled
 from otx.algorithms.detection.adapters.mmdet.hooks.det_saliency_map_hook import (
     DetSaliencyMapHook,
 )
-from otx.mpa.deploy.utils import is_mmdeploy_enabled
 from otx.mpa.modules.hooks.recording_forward_hooks import FeatureVectorHook
 from otx.mpa.modules.utils.task_adapt import map_class_names
 from otx.mpa.utils.logger import get_logger

otx/algorithms/detection/adapters/mmdet/models/detectors/custom_maskrcnn_detector.py

@@ -9,7 +9,7 @@
 from mmdet.models.builder import DETECTORS
 from mmdet.models.detectors.mask_rcnn import MaskRCNN
 
-from otx.mpa.deploy.utils import is_mmdeploy_enabled
+from otx.algorithms.common.adapters.mmdeploy.utils import is_mmdeploy_enabled
 from otx.mpa.modules.hooks.recording_forward_hooks import (
     ActivationMapHook,
     FeatureVectorHook,

otx/algorithms/detection/adapters/mmdet/models/detectors/custom_single_stage_detector.py

@@ -9,10 +9,10 @@
 from mmdet.models.builder import DETECTORS
 from mmdet.models.detectors.single_stage import SingleStageDetector
 
+from otx.algorithms.common.adapters.mmdeploy.utils import is_mmdeploy_enabled
 from otx.algorithms.detection.adapters.mmdet.hooks.det_saliency_map_hook import (
     DetSaliencyMapHook,
 )
-from otx.mpa.deploy.utils import is_mmdeploy_enabled
 from otx.mpa.modules.hooks.recording_forward_hooks import FeatureVectorHook
 from otx.mpa.modules.utils.task_adapt import map_class_names
 from otx.mpa.utils.logger import get_logger

otx/algorithms/detection/adapters/mmdet/models/detectors/custom_yolox_detector.py

@@ -9,10 +9,10 @@
 from mmdet.models.builder import DETECTORS
 from mmdet.models.detectors.yolox import YOLOX
 
+from otx.algorithms.common.adapters.mmdeploy.utils import is_mmdeploy_enabled
 from otx.algorithms.detection.adapters.mmdet.hooks.det_saliency_map_hook import (
     DetSaliencyMapHook,
 )
-from otx.mpa.deploy.utils import is_mmdeploy_enabled
 from otx.mpa.modules.hooks.recording_forward_hooks import FeatureVectorHook
 from otx.mpa.modules.utils.task_adapt import map_class_names
 from otx.mpa.utils.logger import get_logger

otx/algorithms/detection/adapters/mmdet/nncf/patches.py

@@ -19,14 +19,14 @@
 from mmdet.models.roi_heads.bbox_heads.sabl_head import SABLHead
 from mmdet.models.roi_heads.mask_heads.fcn_mask_head import FCNMaskHead
 
+from otx.algorithms.common.adapters.mmdeploy.utils import is_mmdeploy_enabled
 from otx.algorithms.common.adapters.nncf import (
     NNCF_PATCHER,
     is_in_nncf_tracing,
     nncf_trace_wrapper,
     no_nncf_trace_wrapper,
 )
 from otx.algorithms.common.adapters.nncf.patches import nncf_trace_context
-from otx.mpa.deploy.utils import is_mmdeploy_enabled
 
 HEADS_TARGETS = dict(
     classes=(

otx/mpa/cls/exporter.py

@@ -5,7 +5,7 @@
 import numpy as np
 from mmcv.runner import wrap_fp16_model
 
-from otx.mpa.deploy.utils import sync_batchnorm_2_batchnorm
+from otx.algorithms.common.adapters.mmdeploy.utils import sync_batchnorm_2_batchnorm
 from otx.mpa.exporter_mixin import ExporterMixin
 from otx.mpa.registry import STAGES
 from otx.mpa.utils.logger import get_logger
@@ -47,7 +47,7 @@ def model_builder_helper(*args, **kwargs):
     def naive_export(output_dir, model_builder, precision, cfg, model_name="model"):
         from mmcls.datasets.pipelines import Compose
 
-        from ..deploy.apis import NaiveExporter
+        from otx.algorithms.common.adapters.mmdeploy.apis import NaiveExporter
 
         def get_fake_data(cfg, orig_img_shape=(128, 128, 3)):
             pipeline = cfg.data.test.pipeline