yolov5 front-end display, how to perform region detection

I would like to know how to add a region detection function to my code, that is, to detect only a portion of each frame of the video stream and then output it.

    def show_video_frame2(self):
        name_list2 = []

        flag2, img2 = self.cap2.read()
        if img2 is not None:
            showimg2 = img2
            with torch.no_grad():
                img2 = letterbox(img2, new_shape=self.opt.img_size)[0]
                # Convert
                # BGR to RGB, to 3x416x416
                img2 = img2[:, :, ::-1].transpose(2, 0, 1)
                img2 = np.ascontiguousarray(img2)
                img2 = torch.from_numpy(img2).to(self.device)
                img2 = img2.half() if self.half else img2.float()  # uint8 to fp16/32
                img2 /= 255.0  # 0 - 255 to 0.0 - 1.0
                if img2.ndimension() == 3:
                    img2 = img2.unsqueeze(0)
                # Inference
                pred2 = self.model(img2, augment=self.opt.augment)[0]

                # Apply NMS
                pred2 = non_max_suppression(pred2, self.opt.conf_thres, self.opt.iou_thres, classes=self.opt.classes,
                                           agnostic=self.opt.agnostic_nms)
                # Process detections
                for i2, det2 in enumerate(pred2):  # detections per image
                    if det2 is not None and len(det2):
                        # Rescale boxes from img_size to im0 size
                        det2[:, :4] = scale_boxes(
                            img2.shape[2:], det2[:, :4], showimg2.shape).round()
                        # Write results
                        for *xyxy2, conf2, cls2 in reversed(det2):
                            label2 = '%s %.2f' % (self.names[int(cls2)], conf2)
                            name_list2.append(self.names[int(cls2)])
                            print(label2)
                            plot_one_box(
                                xyxy2, showimg2, label=label2, color=self.colors[int(cls2)], line_thickness=2)

            self.out2.write(showimg2)
            show2 = cv2.resize(showimg2, (640, 480))
            self.result2 = cv2.cvtColor(show2, cv2.COLOR_BGR2RGB)
            showImage2 = QtGui.QImage(self.result2.data, self.result2.shape[1], self.result2.shape[0],
                                     QtGui.QImage.Format_RGB888)
            self.label_2.setPixmap(QtGui.QPixmap.fromImage(showImage2))