### 调用YOLOv5的ONNX模型 为了在Python环境中调用YOLOv5的ONNX模型，可以利用`onnxruntime`库来加载并运行该模型。以下是具体实现方法： 安装必要的依赖包[^2]: ```bash pip install onnxruntime numpy opencv-python ``` 编写用于推理的Python脚本如下所示: ```python import cv2 import numpy as np import onnxruntime as ort def preprocess(image, input_size=(640, 640)): ih, iw = image.shape[:2] w, h = input_size scale = min(w / iw, h / ih) nw, nh = int(iw * scale), int(ih * scale) image_resized = cv2.resize(image, (nw, nh)) new_image = np.ones((input_size[1], input_size[0], 3), dtype=np.uint8) * 128 dw = (w - nw) // 2 dh = (h - nh) // 2 new_image[dh:nh + dh, dw:nw + dw, :] = image_resized return new_image def postprocess(outputs, conf_threshold=0.4, iou_threshold=0.5): boxes, scores, class_ids = [], [], [] for output in outputs: box = output[:4] score = output[4] class_id = np.argmax(output[5:]) if score > conf_threshold: boxes.append(box.tolist()) scores.append(score.item()) class_ids.append(class_id.item()) indices = cv2.dnn.NMSBoxes(boxes, scores, conf_threshold, iou_threshold).flatten() filtered_boxes = [boxes[i] for i in indices] filtered_scores = [scores[i] for i in indices] filtered_class_ids = [class_ids[i] for i in indices] return filtered_boxes, filtered_scores, filtered_class_ids if __name__ == "__main__": session = ort.InferenceSession("yolov5s.onnx") img_path = "path_to_test_image.jpg" original_image = cv2.imread(img_path) processed_image = preprocess(original_image) blob = cv2.dnn.blobFromImage(processed_image, 1/255., swapRB=True, crop=False) inputs = {session.get_inputs()[0].name: np.array([blob])} predictions = session.run(None, inputs)[0][0] bboxes, confidences, labels = postprocess(predictions) for bbox, confidence, label in zip(bboxes, confidences, labels): x_center, y_center, width, height = bbox # Adjust bounding box coordinates based on preprocessing steps. ... color = (0, 255, 0) text_color = (0, 0, 255) thickness = 2 font_scale = 0.7 font_face = cv2.FONT_HERSHEY_SIMPLEX cv2.rectangle(original_image, (int(x_center-width//2), int(y_center-height//2)), (int(x_center+width//2), int(y_center+height//2)), color=color, thickness=thickness) cv2.putText(original_image, f"{label}: {confidence:.2f}", org=(int(x_center-width//2), int(y_center-height//2)-10), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=font_scale, color=text_color, thickness=thickness) cv2.imshow('Detected Objects', original_image) cv2.waitKey(0) cv2.destroyAllWindows() ``` 上述代码展示了如何准备输入数据、执行前向传播以及处理输出结果的过程。