minor

Author: Wovchena

model_api/cpp/models/src/detection_model_yolo.cpp

@@ -611,7 +611,7 @@ std::unique_ptr<ResultBase> YoloV8::postprocess(InferenceResult& infResult) {
                 detections[0 * num_proposals + i] - detections[2 * num_proposals + i] / 2.0f,
                 detections[1 * num_proposals + i] - detections[3 * num_proposals + i] / 2.0f,
                 detections[0 * num_proposals + i] + detections[2 * num_proposals + i] / 2.0f,
-                detections[1 * num_proposals + i] + detections[3 * num_proposals + i] / 2.0f
+                detections[1 * num_proposals + i] + detections[3 * num_proposals + i] / 2.0f,
             });
             confidences.push_back(confidence);
             labelIDs.push_back(max_id - 4);  // TODO: move 4 to const

model_api/python/openvino/model_api/adapters/utils.py

@@ -123,16 +123,6 @@ def resize_image_letterbox_graph(input: Output, size, interpolation, pad_value):
         mode=interpolation,
         shape_calculation_mode="sizes",
     )
-    # image = input
-    # image_shape = opset.shape_of(input, name="shape")
-    # nw = opset.convert(
-    #     opset.gather(image_shape, opset.constant(w_axis), axis=0),
-    #     destination_type="i32",
-    # )
-    # nh = opset.convert(
-    #     opset.gather(image_shape, opset.constant(h_axis), axis=0),
-    #     destination_type="i32",
-    # )
     dx = opset.divide(
         opset.subtract(opset.constant(w, dtype=np.int32), nw),
         opset.constant(2, dtype=np.int32),

model_api/python/openvino/model_api/models/image_model.py

@@ -186,20 +186,6 @@ def preprocess(self, inputs):
                 }
             - the input metadata, which might be used in `postprocess` method
         """
-        # import cv2
-        # image = inputs
-        # ih, iw = image.shape[0:2]
-        # w, h = (640, 640)
-        # scale = min(w / iw, h / ih)
-        # nw = round(iw * scale)
-        # nh = round(ih * scale)
-        # image = cv2.resize(image, (nw, nh), interpolation=cv2.INTER_LINEAR)
-
-        # from openvino.model_api.models.utils import resize_image_letterbox
-        # processed = resize_image_letterbox(inputs, (self.w, self.h), cv2.INTER_LINEAR, 114)
-        # processed = (
-        #     processed[None][..., ::-1].transpose((0, 3, 1, 2)).astype(np.float32)
-        # )
         return {self.image_blob_name: inputs[None]}, {
             "original_shape": inputs.shape,
             "resized_shape": (self.w, self.h, self.c),

model_api/python/openvino/model_api/models/model.py

@@ -27,7 +27,7 @@
 
 
 class WrapperError(Exception):
-    """Class for errors occurred in Model API wrappers"""
+    """The class for errors occurred in Model API wrappers"""
 
     def __init__(self, wrapper_name, message):
         super().__init__(f"{wrapper_name}: {message}")

model_api/python/openvino/model_api/models/yolo.py

@@ -748,11 +748,6 @@ def non_max_suppression(
             shape (num_boxes, 6 + num_masks) containing the kept boxes, with columns
             (x1, y1, x2, y2, confidence, class, mask1, mask2, ...).
     """
-    out_shape = prediction.shape
-    if 3 != len(out_shape):
-        raise RuntimeError("YoloV8 wrapper expects the output of rank 3")
-    if 1 != out_shape[0]:
-        raise RuntimeError("YoloV8 wrapper expects 1 as the first dim of the output")
     nc = nc or (prediction.shape[1] - 4)  # number of classes
     mi = 4 + nc  # mask start index
     xc = np.amax(prediction[:, 4:mi], 1) > conf_thres  # candidates
@@ -837,15 +832,19 @@ def parameters(cls):
 
     def postprocess(self, outputs, meta):
         if 1 != len(outputs):
-            raise RuntimeError("YoloV8 wrapper expects 1 output")
+            self.raise_error("expect 1 output")
+        prediction = next(iter(outputs.values()))
+        out_shape = prediction.shape
+        if 3 != len(out_shape):
+            raise RuntimeError("expect the output of rank 3")
+        if 1 != out_shape[0]:
+            raise RuntimeError("expect 1 as the first dim of the output")
         boxes = non_max_suppression(
-            next(iter(outputs.values())), self.confidence_threshold, self.iou_threshold
+            prediction, self.confidence_threshold, self.iou_threshold
         )
 
-        inputImgWidth, inputImgHeight = (
-            meta["original_shape"][1],
-            meta["original_shape"][0],
-        )
+        inputImgWidth = meta["original_shape"][1]
+        inputImgHeight = meta["original_shape"][0]
         invertedScaleX, invertedScaleY = (
             inputImgWidth / self.orig_width,
             inputImgHeight / self.orig_height,
@@ -865,7 +864,7 @@ def postprocess(self, outputs, meta):
         boxes[:, :4] -= (padLeft, padTop, padLeft, padTop)
         boxes[:, :4] *= (invertedScaleX, invertedScaleY, invertedScaleX, invertedScaleY)
 
-        intboxes = np.rint(boxes[:, :4]).astype(np.int32)
+        intboxes = np.rint(boxes[:, :4]).astype(np.int32)  # TODO: np.round(float_act_map, out=float_act_map).astype()
         np.clip(
             intboxes,
             0,

tests/python/accuracy/test_YoloV8.py

@@ -170,7 +170,13 @@ def cached_models(folder, pt):
     )  # TODO: YOLOv5 vs v8
     ref_wrapper = YOLO(export_dir)
     ref_wrapper.overrides["imgsz"] = (impl_wrapper.w, impl_wrapper.h)
-    compiled_model = ov.Core().compile_model(xml, "CPU")
+    if ref_wrapper.predictor is None:
+        ref_wrapper.predict(
+            np.zeros([1, 1, 3], np.uint8)
+        )  # YOLO.predictor is initialized by predict
+    core = ov.Core()
+    ref_wrapper.predictor.model.ov_compiled_model = core.compile_model(ref_wrapper.predictor.model.ov_model, "CPU")
+    compiled_model = core.compile_model(xml, "CPU")
     return impl_wrapper, ref_wrapper, compiled_model
 
 
@@ -290,7 +296,7 @@ def test_detector(impath, data, pt):
     assert (result.boxes.data == ref_predictions.boxes.data).all()
     assert result.boxes.orig_shape == ref_predictions.boxes.orig_shape
     assert result.keypoints == ref_predictions.keypoints
-    assert result.keys == ref_predictions.keys
+    assert result._keys == ref_predictions._keys
     assert result.masks == ref_predictions.masks
     assert result.names == ref_predictions.names
     assert (result.orig_img == ref_predictions.orig_img).all()
@@ -308,7 +314,7 @@ def test_classifier(data):
     ref_wrapper = YOLO(export_path)
     ref_wrapper.overrides["imgsz"] = 224
     im = cv2.imread(data + "/coco128/images/train2017/000000000074.jpg")
-    ref_predictions = ref_wrapper(im)
+    ref_predictions = ref_wrapper.predict(im)
 
     model = ov.Core().compile_model(f"{export_path}/{xmls[0]}")
     orig_imgs = [im]