Apply changes from releases/v0.3.1-geti1.0.0 (#1337)

* anomaly save_model bugfix (#1300) * upgrade networkx module version (#1303) * Forward CVS-94422 size bug fix PR to release branch (#1326) * Valid POT configs for small HRNet models (#1317) * [Release branch] Disable NNCF optimization for FP16 models (#1319) * [RELEASE] CVS-95549 - Hierarchical classification training failed without obvious reason (#1329) * Fix h-label: per-group softmax (#1332) * Fix dataset length bug in mpa task (#1338) * Fix drop_last key issue for det/set (#1340) * Hot-fix for OV inference for iseg output (#1345) * Fix nncf model export bug (#1346) Signed-off-by: Songki Choi <[email protected]> Co-authored-by: Eunwoo Shin <[email protected]> Co-authored-by: Yunchu Lee <[email protected]> Co-authored-by: Eugene Liu <[email protected]> Co-authored-by: Daniil Lyakhov <[email protected]> Co-authored-by: Jihwan Eom <[email protected]> Co-authored-by: Vladisalv Sovrasov <[email protected]> Co-authored-by: Harim Kang <[email protected]> Co-authored-by: Ashwin Vaidya <[email protected]> Co-authored-by: Ashwin Vaidya <[email protected]>
openvinotoolkit · Nov 11, 2022 · 6b199b2 · 6b199b2
1 parent 214401d
commit 6b199b2
Show file tree

Hide file tree

Showing 12 changed files with 34 additions and 29 deletions.
diff --git a/external/anomaly/tasks/inference.py b/external/anomaly/tasks/inference.py
@@ -263,7 +263,7 @@ def export(self, export_type: ExportType, output_model: ModelEntity) -> None:
         output_model.set_data("label_schema.json", label_schema_to_bytes(self.task_environment.label_schema))
         self._set_metadata(output_model)
 
-    def _model_info(self) -> Dict:
+    def model_info(self) -> Dict:
         """Return model info to save the model weights.
 
         Returns:
@@ -282,7 +282,7 @@ def save_model(self, output_model: ModelEntity) -> None:
             output_model (ModelEntity): Output model onto which the weights are saved.
         """
         logger.info("Saving the model weights.")
-        model_info = self._model_info()
+        model_info = self.model_info()
         buffer = io.BytesIO()
         torch.save(model_info, buffer)
         output_model.set_data("weights.pth", buffer.getvalue())
@@ -298,10 +298,13 @@ def save_model(self, output_model: ModelEntity) -> None:
         output_model.optimization_methods = self.optimization_methods
 
     def _set_metadata(self, output_model: ModelEntity):
-        output_model.set_data("image_threshold", self.model.image_threshold.value.cpu().numpy().tobytes())
-        output_model.set_data("pixel_threshold", self.model.pixel_threshold.value.cpu().numpy().tobytes())
-        output_model.set_data("min", self.model.normalization_metrics.state_dict()["min"].cpu().numpy().tobytes())
-        output_model.set_data("max", self.model.normalization_metrics.state_dict()["max"].cpu().numpy().tobytes())
+        if hasattr(self.model, "image_threshold"):
+            output_model.set_data("image_threshold", self.model.image_threshold.value.cpu().numpy().tobytes())
+        if hasattr(self.model, "pixel_threshold"):
+            output_model.set_data("pixel_threshold", self.model.pixel_threshold.value.cpu().numpy().tobytes())
+        if hasattr(self.model, "normalization_metrics"):
+            output_model.set_data("min", self.model.normalization_metrics.state_dict()["min"].cpu().numpy().tobytes())
+            output_model.set_data("max", self.model.normalization_metrics.state_dict()["max"].cpu().numpy().tobytes())
 
     @staticmethod
     def _is_docker() -> bool:

diff --git a/external/anomaly/tasks/nncf.py b/external/anomaly/tasks/nncf.py
@@ -201,7 +201,7 @@ def optimize(
 
         logger.info("Training completed.")
 
-    def _model_info(self) -> Dict:
+    def model_info(self) -> Dict:
         """Return model info to save the model weights.
 
         Returns:

diff --git a/external/mmdetection/detection_tasks/extension/datasets/mmdataset.py b/external/mmdetection/detection_tasks/extension/datasets/mmdataset.py
@@ -139,12 +139,13 @@ def __getitem__(self, index):
 
     @check_input_parameters_type({"ote_dataset": DatasetParamTypeCheck})
     def __init__(
-            self,
-            ote_dataset: DatasetEntity,
-            labels: List[LabelEntity],
-            pipeline: Sequence[dict],
-            domain: Domain,
-            test_mode: bool = False,
+        self,
+        ote_dataset: DatasetEntity,
+        labels: List[LabelEntity],
+        pipeline: Sequence[dict],
+        domain: Domain,
+        test_mode: bool = False,
+        **kwargs,
     ):
         self.ote_dataset = ote_dataset
         self.labels = labels

diff --git a/external/mmsegmentation/segmentation_tasks/extension/datasets/mmdataset.py b/external/mmsegmentation/segmentation_tasks/extension/datasets/mmdataset.py
@@ -82,7 +82,12 @@ class _DataInfoProxy:
         forwards data access operations to ote_dataset and converts the dataset items to the view
         convenient for mmsegmentation.
         """
-        def __init__(self, ote_dataset, labels=None):
+        def __init__(
+            self,
+            ote_dataset,
+            labels=None,
+            **kwargs,
+        ):
             self.ote_dataset = ote_dataset
             self.labels = labels
             self.label_idx = {label.id: i for i, label in enumerate(labels)}

diff --git a/external/mmsegmentation/submodule b/external/mmsegmentation/submodule
diff --git a/external/model-preparation-algorithm/mpa_tasks/apis/classification/task.py b/external/model-preparation-algorithm/mpa_tasks/apis/classification/task.py
@@ -12,7 +12,6 @@
 import torch
 from mmcv.utils import ConfigDict
 from mpa import MPAConstants
-from mpa.stage import Stage
 from mpa.utils.config_utils import MPAConfig
 from mpa.utils.logger import get_logger
 from mpa_tasks.apis import BaseTask, TrainType

diff --git a/external/model-preparation-algorithm/mpa_tasks/apis/task.py b/external/model-preparation-algorithm/mpa_tasks/apis/task.py
@@ -90,6 +90,9 @@ def _run_task(self, stage_module, mode=None, dataset=None, parameters=None, **kw
         self._model_cfg["model_classes"] = model_classes
         if dataset is not None:
             train_data_cfg = Stage.get_data_cfg(self._data_cfg, "train")
+            # if dataset size is smaller than batch size
+            if 0 < len(dataset) < self._recipe_cfg.data.get("samples_per_gpu", 2):
+                train_data_cfg.drop_last = False
             train_data_cfg["data_classes"] = data_classes
             new_classes = np.setdiff1d(data_classes, model_classes).tolist()
             train_data_cfg["new_classes"] = new_classes

diff --git a/external/model-preparation-algorithm/submodule b/external/model-preparation-algorithm/submodule
diff --git a/ote_cli/ote_cli/utils/hpo.py b/ote_cli/ote_cli/utils/hpo.py
@@ -517,12 +517,7 @@ def __init__(
             if _is_anomaly_framework_task(task_type):
                 impl_class = get_impl_class(environment.model_template.entrypoints.base)
                 task = impl_class(task_environment=environment)
-                model = ModelEntity(
-                    dataset,
-                    environment.get_model_configuration(),
-                )
-                task.save_model(model)
-                save_model_data(model, self.work_dir)
+                torch.save(task.model_info(), osp.join(self.work_dir, "weights.pth"))
         else:
             save_model_data(environment.model, self.work_dir)
 

diff --git a/ote_sdk/ote_sdk/usecases/exportable_code/prediction_to_annotation_converter.py b/ote_sdk/ote_sdk/usecases/exportable_code/prediction_to_annotation_converter.py
@@ -181,10 +181,10 @@ def convert_to_annotation(
         image_size = metadata["original_shape"][1::-1]
         for box in predictions:
             scored_label = ScoredLabel(self.labels[int(box.id)], float(box.score))
-            coords = box.get_coords()
+            coords = np.array(box.get_coords(), dtype=float)
             if (coords[2] - coords[0]) * (coords[3] - coords[1]) < 1.0:
                 continue
-            coords = np.array(box.get_coords(), dtype=float) / np.tile(image_size, 2)
+            coords /= np.tile(image_size, 2)
             annotations.append(
                 Annotation(
                     Rectangle(coords[0], coords[1], coords[2], coords[3]),

diff --git a/ote_sdk/ote_sdk/utils/vis_utils.py b/ote_sdk/ote_sdk/utils/vis_utils.py
@@ -14,10 +14,9 @@ def get_actmap(
     saliency_map: np.ndarray,
     output_res: Union[tuple, list],
 ) -> np.ndarray:
-    """
-    Get activation map (heatmap)  from saliency map
+    """Get activation map (heatmap)  from saliency map
 
-    :param saliency_map: Saliency map with pixel values from 0-255 (np.ndarray)
+    :param saliency_map: Saliency map with pixel values from 0-255 (Union[np.ndarray, Iterable, int, float])
     :param output_res: Output resolution (Union[tuple, list])
     :return: activation map, heatmap (np.ndarray)
     """

diff --git a/ote_sdk/requirements.txt b/ote_sdk/requirements.txt
@@ -1,7 +1,7 @@
 numpy>=1.16.4
 scikit-learn==0.24.*
 Shapely>=1.7.1,<=1.8.0
-networkx>=2.5,<2.8.1rc1
+networkx>=2.6,<2.8.1rc1
 opencv-python==4.5.*
 pymongo==3.12.0
 omegaconf==2.1.*
+2 −1		mmseg/core/evaluation/metrics.py
+5 −1		mmseg/datasets/pipelines/loading.py