yolov8改进损失函数DIOU

### 实现 DIoU 损失函数改进版本 为了在 YOLOv8 中引入 DIoU (Distance-IoU) 损失函数，需对原有损失计算部分进行调整。DIoU 是一种基于 IoU 的改进型损失函数，在传统 IoU 基础上加入了中心点距离惩罚项，有助于提升模型定位准确性。 #### 修改思路 通过分析现有 MPDIoU 改进方案[^1] 和其他相关研究进展[^2]，可以在 `bbox_iou` 函数中加入 DIoU 计算逻辑： ```python def bbox_iou(box1, box2, DIOU=True, x1y1x2y2=False): """ Returns the IoU of two bounding boxes along with distance component for DIoU. Args: box1: First set of boxes. box2: Second set of boxes. DIOU: Flag to enable/disable DIoU computation. x1y1x2y2: Boolean flag indicating whether coordinates are given as (x_min, y_min, x_max, y_max). Returns: Tensor containing IoUs or DIoUs between each pair of boxes from both sets. """ # Get the coordinates of bounding boxes if not x1y1x2y2: b1_x1, b1_y1, b1_x2, b1_y2 = box1[:, 0], box1[:, 1], box1[:, 2], box1[:, 3] b2_x1, b2_y1, b2_x2, b2_y2 = box2[:, 0], box2[:, 1], box2[:, 2], box2[:, 3] else: # Transform from center and width to exact coordinates b1_x1, b1_x2 = box1[:, 0] - box1[:, 2] / 2, box1[:, 0] + box1[:, 2] / 2 b1_y1, b1_y2 = box1[:, 1] - box1[:, 3] / 2, box1[:, 1] + box1[:, 3] / 2 b2_x1, b2_x2 = box2[:, 0] - box2[:, 2] / 2, box2[:, 0] + box2[:, 2] / 2 b2_y1, b2_y2 = box2[:, 1] - box2[:, 3] / 2, box2[:, 1] + box2[:, 3] / 2 # Intersection area inter_area = torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1) inter_area *= torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1) # Union Area w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 union_area = (w1 * h1 + 1e-16) + w2 * h2 - inter_area iou = inter_area / union_area if DIOU: cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex width ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height c2 = cw ** 2 + ch ** 2 + 1e-16 rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 + (b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4 # center dist squared diou_loss = iou - rho2 / c2 return diou_loss.clamp(min=-1.0, max=1.0).mean() return iou.mean() ``` 此代码片段展示了如何扩展原始的 IoU 计算方法来支持 DIoU 损失计算。当参数 `DIOU=True` 时，除了常规交并比外还会考虑两个边界框中心之间的相对位置关系，从而形成更加鲁棒的目标检测性能评估指标[^3]。