[POT] Update samples and samplers with the new DataLoader format (#10595)

* Update samples and samplers with the new DataLoader format

* Update with utils

* Pylint updates

* Update metric with the exception

* Pylint

* Update with the exception

* Pylint

* Revert index sampler changes

* Update ImageLoader & SimplifiedEngine

* Update with the different solution

* Remove utils

* Pylint

* Remove list wrapping

* Remove list from meta_data

tools/pot/openvino/tools/pot/api/metric.py

@@ -12,9 +12,10 @@ class Metric(ABC):
         self.reset()
 
     @property
-    @abstractmethod
     def value(self):
         """ Returns accuracy metric value for the last model output. """
+        raise Exception('The value() property should be implemented to use this metric '
+                        'with AccuracyAwareQuantization algorithm!')
 
     @property
     @abstractmethod

tools/pot/openvino/tools/pot/api/samples/3d_segmentation/3d_segmentation_sample.py

@@ -38,9 +38,8 @@ class BRATSDataLoader(DataLoader):
         mask_path = os.path.join(self.config.mask_dir, self._img_ids[index])
         image_path = os.path.join(self.config.data_source, self._img_ids[index])
 
-        annotation = (index, self._read_image(mask_path))
         image, image_meta = self._preprocess_image(self._read_image(image_path))
-        return annotation, image, image_meta
+        return image, self._read_image(mask_path), image_meta
 
     def __len__(self):
         """ Returns size of the dataset """
@@ -120,13 +119,6 @@ class DiceIndex(Metric):
         self._name = 'dice_index'
         self._overall_metric = []
 
-    @property
-    def value(self):
-        """ Returns accuracy metric value for the last model output.
-        Possible format: {metric_name: [metric_values_per_image]}
-        """
-        return {self._name: [np.mean(self._overall_metric[-1])]}
-
     @property
     def avg_value(self):
         """ Returns accuracy metric value for all model outputs.
@@ -195,6 +187,7 @@ class SegmentationEngine(IEEngine):
         processed_outputs = []
         for output, meta in zip(outputs.values(), metadata):
             # Resize to bounding box size and extend to mask size
+            output = output[0]
             low = meta['bbox'][0]
             high = meta['bbox'][1]
             box_shape = tuple((high - low).astype(np.int32))

tools/pot/openvino/tools/pot/api/samples/classification/classification_sample.py

@@ -29,9 +29,8 @@ class ImageNetDataLoader(DataLoader):
         if index >= len(self):
             raise IndexError
 
-        annotation = (index, self._annotations[self._img_ids[index]])\
+        annotation = self._annotations[self._img_ids[index]] if self._annotations else None
-            if self._annotations else (index, None)
+        return self._read_image(self._img_ids[index]), annotation
-        return annotation, self._read_image(self._img_ids[index])
 
     # Methods specific to the current implementation
     @staticmethod
@@ -84,11 +83,6 @@ class Accuracy(Metric):
         self._name = 'accuracy@top{}'.format(self._top_k)
         self._matches = []
 
-    @property
-    def value(self):
-        """ Returns accuracy metric value for the last model output. """
-        return {self._name: self._matches[-1]}
-
     @property
     def avg_value(self):
         """ Returns accuracy metric value for all model outputs. """

tools/pot/openvino/tools/pot/api/samples/face_detection/face_detection_sample.py

@@ -46,8 +46,7 @@ class WiderFaceLoader(DataLoader):
         if index >= len(self):
             raise IndexError
 
-        annotation = (index, self._annotations[self._img_ids[index]])
+        return self._read_image(self._img_ids[index]), self._annotations[self._img_ids[index]]
-        return annotation, self._read_image(self._img_ids[index])
 
     def __len__(self):
         """ Returns size of the dataset """
@@ -312,15 +311,6 @@ class Recall(Metric):
         self._n_recorded_faces = []
         self._n_total_preds = []
 
-    @property
-    def value(self):
-        """ Returns metric value for the last model output.
-         Possible format: {metric_name: [metric_values_per_image]}
-         """
-        tp = np.cumsum(self._true_positives[-1])[np.arange(self._n_total_preds[-1])]
-        recalls = tp / np.maximum(self._n_recorded_faces[-1], np.finfo(np.float64).eps)
-        return {self._name: [recalls[-1]]}
-
     @property
     def avg_value(self):
         """ Returns average metric value for all model outputs.

tools/pot/openvino/tools/pot/api/samples/object_detection/data_loader.py

@@ -12,8 +12,8 @@ from openvino.tools.pot import DataLoader
 class COCOLoader(DataLoader):
     def __init__(self, config):
         super().__init__(config)
-        self.images_path = config.images_path
+        self.images_path = self.config.images_path
-        self.annotation_path = config.annotation_path
+        self.annotation_path = self.config.annotation_path
         self.images = os.listdir(self.images_path)
         self.labels = None
         self.data, self.bbox = self.prepare_annotation()
@@ -61,8 +61,8 @@ class COCOLoader(DataLoader):
 
         annotation = {'boxes': bbox, 'labels': labels, 'iscrowd': iscrowd,
                       'x_maxs': x_maxs, 'x_mins': x_mins, 'y_maxs': y_maxs, 'y_mins': y_mins}
-        annotation = (index, [annotation, shape_image])
+        annotation = [annotation, shape_image]
-        return annotation, self._read_and_preprocess_image(self.images_path + self.data[index]['file_name'])
+        return self._read_and_preprocess_image(self.images_path + self.data[index]['file_name']), annotation
 
     def __len__(self):
         return len(self.images)

tools/pot/openvino/tools/pot/api/samples/segmentation/segmentation_sample.py

@@ -33,8 +33,8 @@ class VOCSegmentationLoader(DataLoader):
     # Required methods:
     def __init__(self, config):
         super().__init__(config)
-        self._image_size = config.image_size
+        self._image_size = self.config.image_size
-        self._img_ids = self._read_img_ids(config)
+        self._img_ids = self._read_img_ids(self.config)
 
     def __getitem__(self, index):
         """
@@ -49,8 +49,7 @@ class VOCSegmentationLoader(DataLoader):
         mask_path = os.path.join(self.config.mask_dir, self._img_ids[index] + '.png')
         image_path = os.path.join(self.config.data_source, self._img_ids[index] + '.jpg')
 
-        annotation = (index, self._read_and_preprocess_mask(mask_path))
+        return self._read_and_preprocess_image(image_path), self._read_and_preprocess_mask(mask_path)
-        return annotation, self._read_and_preprocess_image(image_path)
 
     def __len__(self):
         """ Returns size of the dataset """
@@ -93,13 +92,6 @@ class MeanIOU(Metric):
         self._current_cm = []
         self._total_cm = np.zeros((self._classes_num, self._classes_num))
 
-    @property
-    def value(self):
-        """ Returns metric value for the last model output.
-         Possible format: {metric_name: [metric_values_per_image]}
-         """
-        return {self._name: [self._evaluate(cm) for cm in self._current_cm]}
-
     @property
     def avg_value(self):
         """ Returns average metric value for all model outputs.

tools/pot/openvino/tools/pot/engines/ie_engine.py

@@ -403,11 +403,15 @@ class IEEngine(Engine):
             raise RuntimeError('Inconsistent data in the batch. '
                                'Some items contain annotation, and some do not.')
 
+        if not all([isinstance(item[0], tuple) for item in batch]):
+            images, image_annotation = [data[0] for data in batch], [(idx, data[1]) for idx, data in enumerate(batch)]
+        else:
+            images, image_annotation = [data[1] for data in batch], [data[0] for data in batch]
+
         if all([len(item) == 2 for item in batch]):
-            image_annotation, images = map(list, zip(*batch))
             meta_data = [{}]*len(images)
         elif all([len(item) == 3 for item in batch]):
-            image_annotation, images, meta_data = map(list, zip(*batch))
+            meta_data = [data[2] for data in batch]
         else:
             raise RuntimeError('Inconsistent data in the batch. '
                                'Some items contain meta data, and some do not.')