AC: Added support for ssd-resnet-34-tf model (openvinotoolkit#3009)

* Added support for ssd-resnet-34-tf model

* Updated readme

Author: Anna Grebneva

tools/accuracy_checker/openvino/tools/accuracy_checker/adapters/README.md

@@ -489,4 +489,6 @@ AccuracyChecker supports following set of adapters:
   * `boxes_out` - name of output layer with bounding boxes coordinates.
   * `confidence_threshold` - lower bound for valid boxes scores (optional, default 0.01).
   * `nms_threshold` - overlap threshold for NMS (optional, default 0.45).
-  * `keep_top_k ` - maximal number of boxes which should be kept (optional, default 200).
+  * `keep_top_k` - maximal number of boxes which should be kept during NMS (optional, default 200).
+  * `diff_coord_order` - ordering convention of coordinates differs from the commonly used format [x0, y0, x1, y1]. If value is True, the format of coordinates is [y0, x0, y1, x1] (optional, default False).
+  * `max_detections` - maximal number of boxes which should be kept (optional).

tools/accuracy_checker/openvino/tools/accuracy_checker/adapters/ssd.py

@@ -393,7 +393,10 @@ def parameters(cls):
             'boxes_out': StringField(description='boxes output'),
             'confidence_threshold': NumberField(optional=True, default=0.01, description="confidence threshold"),
             'nms_threshold': NumberField(optional=True, default=0.45, description="NMS threshold"),
-            'keep_top_k': NumberField(optional=True, value_type=int, default=200, description="keep top K")
+            'keep_top_k': NumberField(optional=True, value_type=int, default=200, description="keep top K during NMS"),
+            'diff_coord_order': BoolField(optional=True, default=False,
+                                          description='Ordering convention of coordinates differs from standard'),
+            'max_detections': NumberField(optional=True, value_type=int, description="maximal number of detections")
         })
         return params
 
@@ -403,6 +406,8 @@ def configure(self):
         self.confidence_threshold = self.get_value_from_config('confidence_threshold')
         self.iou_threshold = self.get_value_from_config('nms_threshold')
         self.keep_top_k = self.get_value_from_config('keep_top_k')
+        self.diff_coord_order = self.get_value_from_config('diff_coord_order')
+        self.max_detections = self.get_value_from_config('max_detections')
         self.outputs_verified = False
 
     def select_output_blob(self, outputs):
@@ -426,7 +431,10 @@ def process(self, raw, identifiers, frame_meta):
                     continue
                 subset_boxes = boxes[mask, :]
 
-                x_mins, y_mins, x_maxs, y_maxs = subset_boxes.T
+                if self.diff_coord_order:
+                    y_mins, x_mins, y_maxs, x_maxs = subset_boxes.T
+                else:
+                    x_mins, y_mins, x_maxs, y_maxs = subset_boxes.T
 
                 keep = NMS.nms(x_mins, y_mins, x_maxs, y_maxs, probs, self.iou_threshold, include_boundaries=False,
                                keep_top_k=self.keep_top_k)
@@ -445,6 +453,15 @@ def process(self, raw, identifiers, frame_meta):
                 detections['x_maxs'].extend(x_maxs)
                 detections['y_maxs'].extend(y_maxs)
 
+            if self.max_detections:
+                max_ids = np.argsort(-np.array(detections['scores']))[:self.max_detections]
+                detections['labels'] = np.array(detections['labels'])[max_ids]
+                detections['scores'] = np.array(detections['scores'])[max_ids]
+                detections['x_mins'] = np.array(detections['x_mins'])[max_ids]
+                detections['y_mins'] = np.array(detections['y_mins'])[max_ids]
+                detections['x_maxs'] = np.array(detections['x_maxs'])[max_ids]
+                detections['y_maxs'] = np.array(detections['y_maxs'])[max_ids]
+
             result.append(DetectionPrediction(identifier, detections['labels'], detections['scores'],
                                               detections['x_mins'], detections['y_mins'], detections['x_maxs'],
                                               detections['y_maxs']))