超越CIOU/SIOU | Wise-IoU助力YOLO强势涨点！！！

论文题目：Wise-IoU: Bounding Box Regression Loss with Dynamic Focusing Mechanism

论文链接：https://arxiv.org/abs/2301.10051

近年来的研究大多假设训练数据中的示例有较高的质量，致力于强化边界框损失的拟合能力。但注意到目标检测训练集中含有低质量示例，如果一味地强化边界框对低质量示例的回归，显然会危害模型检测性能的提升。

Focal-EIoU v1 被提出以解决这个问题，但由于其聚焦机制是静态的，并未充分挖掘非单调聚焦机制的潜能。基于这个观点，作者提出了动态非单调的聚焦机制，设计了 Wise-IoU (WIoU)。动态非单调聚焦机制使用“离群度”替代 IoU 对锚框进行质量评估，并提供了明智的梯度增益分配策略。该策略在降低高质量锚框的竞争力的同时，也减小了低质量示例产生的有害梯度。这使得 WIoU 可以聚焦于普通质量的锚框，并提高检测器的整体性能。将WIoU应用于最先进的单级检测器 YOLOv7 时，在 MS-COCO 数据集上的 AP-75 从 53.03% 提升到 54.50%。

一、 Wise-IoU相关代码

import mathimport torchclass IoU_Cal:''' pred, target: x0,y0,x1,y1monotonous: {None: originTrue: monotonic FMFalse: non-monotonic FM}momentum: The momentum of running mean (This can be set by the function <momentum_estimation>)'''iou_mean = 1.monotonous = Falsemomentum = 1 - pow(0.05, 1 / (890 * 34))_is_train = True@classmethoddef momentum_estimation(cls, n, t):''' n: Number of batches per training epocht: The epoch when mAP's ascension slowed significantly'''time_to_real = n * tcls.momentum = 1 - pow(0.05, 1 / time_to_real)return cls.momentumdef __init__(self, pred, target):self.pred, self.target = pred, targetself._fget = {# x,y,w,h'pred_xy': lambda: (self.pred[..., :2] + self.pred[..., 2: 4]) / 2,'pred_wh': lambda: self.pred[..., 2: 4] - self.pred[..., :2],'target_xy': lambda: (self.target[..., :2] + self.target[..., 2: 4]) / 2,'target_wh': lambda: self.target[..., 2: 4] - self.target[..., :2],# x0,y0,x1,y1'min_coord': lambda: torch.minimum(self.pred[..., :4], self.target[..., :4]),'max_coord': lambda: torch.maximum(self.pred[..., :4], self.target[..., :4]),# The overlapping region'wh_inter': lambda: torch.relu(self.min_coord[..., 2: 4] - self.max_coord[..., :2]),'s_inter': lambda: torch.prod(self.wh_inter, dim=-1),# The area covered's_union': lambda: torch.prod(self.pred_wh, dim=-1) +torch.prod(self.target_wh, dim=-1) - self.s_inter,# The smallest enclosing box'wh_box': lambda: self.max_coord[..., 2: 4] - self.min_coord[..., :2],'s_box': lambda: torch.prod(self.wh_box, dim=-1),'l2_box': lambda: torch.square(self.wh_box).sum(dim=-1),# The central points' connection of the bounding boxes'd_center': lambda: self.pred_xy - self.target_xy,'l2_center': lambda: torch.square(self.d_center).sum(dim=-1),# IoU'iou': lambda: 1 - self.s_inter / self.s_union}self._update(self)def __setitem__(self, key, value):self._fget[key] = valuedef __getattr__(self, item):if callable(self._fget[item]):self._fget[item] = self._fget[item]()return self._fget[item]@classmethoddef train(cls):cls._is_train = True@classmethoddef eval(cls):cls._is_train = False@classmethoddef _update(cls, self):if cls._is_train: cls.iou_mean = (1 - cls.momentum) * cls.iou_mean + \cls.momentum * self.iou.detach().mean().item()def _scaled_loss(self, loss, alpha=1.9, delta=3):if isinstance(self.monotonous, bool):beta = self.iou.detach() / self.iou_meanif self.monotonous:loss *= beta.sqrt()else:divisor = delta * torch.pow(alpha, beta - delta)loss *= beta / divisorreturn loss@classmethoddef IoU(cls, pred, target, self=None):self = self if self else cls(pred, target)return self.iou@classmethoddef WIoU(cls, pred, target, self=None):self = self if self else cls(pred, target)dist = torch.exp(self.l2_center / self.l2_box.detach())return self._scaled_loss(dist * self.iou)@classmethoddef EIoU(cls, pred, target, self=None):self = self if self else cls(pred, target)penalty = self.l2_center / self.l2_box.detach() \+ torch.square(self.d_center / self.wh_box).sum(dim=-1)return self._scaled_loss(self.iou + penalty)@classmethoddef GIoU(cls, pred, target, self=None):self = self if self else cls(pred, target)return self._scaled_loss(self.iou + (self.s_box - self.s_union) / self.s_box)@classmethoddef DIoU(cls, pred, target, self=None):self = self if self else cls(pred, target)return self._scaled_loss(self.iou + self.l2_center / self.l2_box)@classmethoddef CIoU(cls, pred, target, eps=1e-4, self=None):self = self if self else cls(pred, target)v = 4 / math.pi ** 2 * \(torch.atan(self.pred_wh[..., 0] / (self.pred_wh[..., 1] + eps)) -torch.atan(self.target_wh[..., 0] / (self.target_wh[..., 1] + eps))) ** 2alpha = v / (self.iou + v)return self._scaled_loss(self.iou + self.l2_center / self.l2_box + alpha.detach() * v)@classmethoddef SIoU(cls, pred, target, theta=4, self=None):self = self if self else cls(pred, target)# Angle Costangle = torch.arcsin(torch.abs(self.d_center).min(dim=-1)[0] / (self.l2_center.sqrt() + 1e-4))angle = torch.sin(2 * angle) - 2# Dist Costdist = angle[..., None] * torch.square(self.d_center / self.wh_box)dist = 2 - torch.exp(dist[..., 0]) - torch.exp(dist[..., 1])# Shape Costd_shape = torch.abs(self.pred_wh - self.target_wh)big_shape = torch.maximum(self.pred_wh, self.target_wh)w_shape = 1 - torch.exp(- d_shape[..., 0] / big_shape[..., 0])h_shape = 1 - torch.exp(- d_shape[..., 1] / big_shape[..., 1])shape = w_shape ** theta + h_shape ** thetareturn self._scaled_loss(self.iou + (dist + shape) / 2)