rename threshold parameter (#2964)

Author: Aleksei Korobeinikov

demos/common/python/openvino/model_zoo/model_api/models/centernet.py

@@ -61,7 +61,7 @@ def postprocess(self, outputs, meta):
                                  xs + wh[..., 0:1] / 2,
                                  ys + wh[..., 1:2] / 2), axis=1)
         detections = np.concatenate((bboxes, scores, clses), axis=1)
-        mask = detections[..., 4] >= self.threshold
+        mask = detections[..., 4] >= self.confidence_threshold
         filtered_detections = detections[mask]
         scale = max(meta['original_shape'])
         center = np.array(meta['original_shape'][:2])/2.0

demos/common/python/openvino/model_zoo/model_api/models/ctpn.py

@@ -79,7 +79,7 @@ def parameters(cls):
             'iou_threshold': NumericalValue(default_value=0.5, description="Threshold for NMS filtering"),
             'input_size': ListValue()
         })
-        parameters['threshold'].update_default_value(0.9)
+        parameters['confidence_threshold'].update_default_value(0.9)
         parameters['labels'].update_default_value(['Text'])
         return parameters
 
@@ -207,7 +207,7 @@ def get_detections(self, text_proposals, scores, size):
         heights = (abs(text_recs[:, 5] - text_recs[:, 1]) + abs(text_recs[:, 7] - text_recs[:, 3])) / 2.0 + 1
         widths = (abs(text_recs[:, 2] - text_recs[:, 0]) + abs(text_recs[:, 6] - text_recs[:, 4])) / 2.0 + 1
         scores = text_recs[:, 8]
-        keep_inds = np.where((widths / heights > self.min_ratio) & (scores > self.threshold) &
+        keep_inds = np.where((widths / heights > self.min_ratio) & (scores > self.confidence_threshold) &
                              (widths > self.min_width))[0]
 
         return text_recs[keep_inds]

demos/common/python/openvino/model_zoo/model_api/models/detection_model.py

@@ -57,7 +57,7 @@ def __init__(self, model_adapter, configuration=None, preload=False):
     def parameters(cls):
         parameters = super().parameters()
         parameters.update({
-            'threshold': NumericalValue(default_value=0.5, description="Threshold value for detection box confidence"),
+            'confidence_threshold': NumericalValue(default_value=0.5, description="Threshold value for detection box confidence"),
             'labels': ListValue(description="List of class labels"),
             'path_to_labels': StringValue(
                 description="Path to file with labels. Overrides the labels, if they sets via 'labels' parameter"

demos/common/python/openvino/model_zoo/model_api/models/detr.py

@@ -47,7 +47,7 @@ def _get_outputs(self):
     def parameters(cls):
         parameters = super().parameters()
         parameters['resize_type'].update_default_value('standard')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         return parameters
 
     def postprocess(self, outputs, meta):
@@ -65,7 +65,7 @@ def _parse_outputs(self, outputs):
         labels = np.argmax(scores[:, :-1], axis=-1)
         det_scores = np.max(scores[:, :-1], axis=-1)
 
-        keep = det_scores > self.threshold
+        keep = det_scores > self.confidence_threshold
 
         detections = [Detection(*det) for det in zip(x_mins[keep], y_mins[keep], x_maxs[keep], y_maxs[keep],
                                                      det_scores[keep], labels[keep])]

demos/common/python/openvino/model_zoo/model_api/models/faceboxes.py

@@ -75,7 +75,7 @@ def _parse_outputs(self, outputs, meta):
 
         score = np.transpose(scores)[1]
 
-        mask = score > self.threshold
+        mask = score > self.confidence_threshold
         filtered_boxes, filtered_score = boxes[mask, :], score[mask]
         if filtered_score.size != 0:
             x_mins = (filtered_boxes[:, 0] - 0.5 * filtered_boxes[:, 2])

demos/common/python/openvino/model_zoo/model_api/models/retinaface.py

@@ -43,14 +43,14 @@ def __init__(self, model_adapter, configuration=None, preload=False):
     def parameters(cls):
         parameters = super().parameters()
         parameters['resize_type'].update_default_value('standard')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         return parameters
 
     def postprocess(self, outputs, meta):
         scale_x = meta['resized_shape'][1] / meta['original_shape'][1]
         scale_y = meta['resized_shape'][0] / meta['original_shape'][0]
 
-        outputs = self.postprocessor.process_output(outputs, scale_x, scale_y, self.threshold, self.mask_threshold)
+        outputs = self.postprocessor.process_output(outputs, scale_x, scale_y, self.confidence_threshold, self.mask_threshold)
         return clip_detections(outputs, meta['original_shape'])
 
 
@@ -69,15 +69,15 @@ def __init__(self, model_adapter, configuration=None, preload=False):
     def parameters(cls):
         parameters = super().parameters()
         parameters['resize_type'].update_default_value('standard')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         parameters['labels'].update_default_value(['Face'])
         return parameters
 
     def postprocess(self, outputs, meta):
         scale_x = meta['resized_shape'][1] / meta['original_shape'][1]
         scale_y = meta['resized_shape'][0] / meta['original_shape'][0]
 
-        outputs = self.postprocessor.process_output(outputs, scale_x, scale_y, self.threshold,
+        outputs = self.postprocessor.process_output(outputs, scale_x, scale_y, self.confidence_threshold,
                                                     meta['resized_shape'][:2])
         return clip_detections(outputs, meta['original_shape'])
 

demos/common/python/openvino/model_zoo/model_api/models/ssd.py

@@ -32,7 +32,7 @@ def __init__(self, model_adapter, configuration=None, preload=False):
     def parameters(cls):
         parameters = super().parameters()
         parameters['resize_type'].update_default_value('standard')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         return parameters
 
     def preprocess(self, inputs):
@@ -72,7 +72,7 @@ def _get_output_parser(self, image_blob_name, bboxes='bboxes', labels='labels',
     def _parse_outputs(self, outputs, meta):
         detections = self.output_parser(outputs)
 
-        detections = [d for d in detections if d.score > self.threshold]
+        detections = [d for d in detections if d.score > self.confidence_threshold]
 
         return detections
 

demos/common/python/openvino/model_zoo/model_api/models/ultra_lightweight_face_detection.py

@@ -54,7 +54,7 @@ def parameters(cls):
             'iou_threshold': NumericalValue(default_value=0.5, description="Threshold for NMS filtering"),
         })
         parameters['resize_type'].update_default_value('standard')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         parameters['labels'].update_default_value(['Face'])
         return parameters
 
@@ -69,7 +69,7 @@ def _parse_outputs(self, outputs, meta):
 
         score = np.transpose(scores)[1]
 
-        mask = score > self.threshold
+        mask = score > self.confidence_threshold
         filtered_boxes, filtered_score = boxes[mask, :], score[mask]
 
         x_mins, y_mins, x_maxs, y_maxs = filtered_boxes.T

demos/common/python/openvino/model_zoo/model_api/models/yolo.py

@@ -115,7 +115,7 @@ def parameters(cls):
             'iou_threshold': NumericalValue(default_value=0.5, description="Threshold for NMS filtering"),
         })
         parameters['resize_type'].update_default_value('fit_to_window_letterbox')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         return parameters
 
     def postprocess(self, outputs, meta):
@@ -124,7 +124,7 @@ def postprocess(self, outputs, meta):
         return detections
 
     @staticmethod
-    def _parse_yolo_region(cls, predictions, input_size, params, threshold):
+    def _parse_yolo_region(cls, predictions, input_size, params, confidence_threshold):
         # ------------------------------------------ Extracting layer parameters ---------------------------------------
         objects = []
         size_normalizer = input_size if params.use_input_size else params.sides
@@ -135,7 +135,7 @@ def _parse_yolo_region(cls, predictions, input_size, params, threshold):
             class_probabilities = cls._get_probabilities(prediction, params.classes)
 
             # filter out the proposals with low confidence score
-            keep_idxs = np.nonzero(class_probabilities > threshold)[0]
+            keep_idxs = np.nonzero(class_probabilities > confidence_threshold)[0]
             class_probabilities = class_probabilities[keep_idxs]
             obj_indx = keep_idxs // params.classes
             class_idx = keep_idxs % params.classes
@@ -222,7 +222,7 @@ def _parse_outputs(self, outputs, meta):
             out_blob = outputs[layer_name]
             layer_params = self.yolo_layer_params[layer_name]
             out_blob.shape = layer_params[0]
-            detections += self._parse_yolo_region(self, out_blob, meta['resized_shape'], layer_params[1], self.threshold)
+            detections += self._parse_yolo_region(self, out_blob, meta['resized_shape'], layer_params[1], self.confidence_threshold)
 
         detections = self._filter(detections, self.iou_threshold)
         return detections
@@ -358,7 +358,7 @@ def parameters(cls):
         parameters.update({
             'iou_threshold': NumericalValue(default_value=0.65, description="Threshold for NMS filtering"),
         })
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         return parameters
 
     def preprocess(self, inputs):
@@ -385,11 +385,11 @@ def postprocess(self, outputs, meta):
             output[..., :2] = (output[..., :2] + self.grids) * self.expanded_strides
             output[..., 2:4] = np.exp(output[..., 2:4]) * self.expanded_strides
 
-        valid_predictions = output[output[..., 4] > self.threshold]
+        valid_predictions = output[output[..., 4] > self.confidence_threshold]
         valid_predictions[:, 5:] *= valid_predictions[:, 4:5]
 
         boxes = self.xywh2xyxy(valid_predictions[:, :4]) / meta['scale']
-        i, j = (valid_predictions[:, 5:] > self.threshold).nonzero()
+        i, j = (valid_predictions[:, 5:] > self.confidence_threshold).nonzero()
         x_mins, y_mins, x_maxs, y_maxs = boxes[i].T
         scores = valid_predictions[i, j + 5]
 
@@ -463,7 +463,7 @@ def _get_outputs(self):
     def parameters(cls):
         parameters = super().parameters()
         parameters['resize_type'].update_default_value('fit_to_window_letterbox')
-        parameters['threshold'].update_default_value(0.5)
+        parameters['confidence_threshold'].update_default_value(0.5)
         return parameters
 
     def preprocess(self, inputs):
@@ -497,7 +497,7 @@ def _parse_outputs(self, outputs):
             out_scores.append(scores[tuple(idx_[1:])])
             out_boxes.append(boxes[idx_[2]])
         transposed_boxes = np.array(out_boxes).T if out_boxes else ([], [], [], [])
-        mask = np.array(out_scores) > self.threshold
+        mask = np.array(out_scores) > self.confidence_threshold
 
         if mask.size == 0:
             return []

demos/object_detection_demo/python/object_detection_demo.py

@@ -217,7 +217,7 @@ def main():
         'scale_values': args.scale_values,
         'reverse_input_channels': args.reverse_input_channels,
         'path_to_labels': args.labels,
-        'threshold': args.prob_threshold,
+        'confidence_threshold': args.prob_threshold,
         'input_size': args.input_size, # The CTPN specific
     }
     model = DetectionModel.create_model(args.architecture_type, model_adapter, configuration)