Boost the rising point | yolov5 combined with alpha IOU

Thesis title ：《Alpha-IoU: A Family of Power Intersection over Union Losses for Bounding Box Regression》

Address of thesis ：  https://arxiv.org/abs/2110.13675v2

1. Brief introduction of the paper ：

In this paper, , The author bases the existing on IoU Loss To a new Power IoU series Loss, The series has a Power IoU Item and an additional Power The regularization , Have a single Power Parameters α. Call this new loss series α-IoU Loss. Experiments on multi-target detection benchmarks and models show that ,α-IoU Loss ：

• It can significantly exceed the existing IoU The loss of ;

• By adjusting the α, Enable the detector to achieve different levels of bbox Greater flexibility in regression accuracy ;

• It is more robust to small data sets and noise .

Experimental results show that ,α(α>1) Added high IoU Target loss and gradient , And then improve bbox Regression accuracy .

power Parameters α It can be used as an adjustment α-IoU Loss of superparameters to meet different levels of bbox Regression accuracy , among α >1 By paying more attention to High IoU Objective to obtain high regression accuracy ( namely High IoU threshold ).

α Not overly sensitive to different models or data sets , in the majority of cases ,α=3 Always perform well .α-IoU The loss family can be easily used to improve the effect of the detector , In a clean or noisy environment , No additional parameters will be introduced , No more training / Reasoning time .

2. Corresponding code ：

def bbox_alpha_iou(box1, box2, x1y1x2y2=False, GIoU=False, DIoU=False, CIoU=False, EIoU=False, alpha=3, eps=1e-9):
    # Returns tsqrt_he IoU of box1 to box2. box1 is 4, box2 is nx4
    box2 = box2.T
 
    # Get the coordinates of bounding boxes
    if x1y1x2y2:  # x1, y1, x2, y2 = box1
        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 xywh to xyxy
        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 = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
            (torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
 
    # Union Area
    w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
    w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps
    union = w1 * h1 + w2 * h2 - inter + eps
 
    # change iou into pow(iou+eps)  Join in α The next power 
    # alpha iou
    iou = torch.pow(inter / union + eps, alpha)
    beta = 2 * alpha
    if GIoU or DIoU or CIoU or EIoU:
        #  The minimum closure region of two boxes width and height
        cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1)  # convex (smallest enclosing box) width
        ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1)  # convex height
 
        if CIoU or DIoU or EIoU:  # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
            #  Minimum circumscribed rectangle   The length of the diagonal is squared 
            c2 = cw ** beta + ch ** beta + eps  # convex diagonal
            rho_x = torch.abs(b2_x1 + b2_x2 - b1_x1 - b1_x2)
            rho_y = torch.abs(b2_y1 + b2_y2 - b1_y1 - b1_y2)
            #  The square of the distance between the center points of two frames 
            rho2 = (rho_x ** beta + rho_y ** beta) / (2 ** beta)  # center distance
            if DIoU:
                return iou - rho2 / c2  # DIoU
 
            elif CIoU:  # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
                v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
                with torch.no_grad():
                    alpha_ciou = v / ((1 + eps) - inter / union + v)
                # return iou - (rho2 / c2 + v * alpha_ciou)  # CIoU
                return iou - (rho2 / c2 + torch.pow(v * alpha_ciou + eps, alpha))  # CIoU
 
            # EIoU  stay CIoU On the basis of 
            #  The aspect ratio loss term of the predicted frame width and height   The difference between the width and height of the split prediction frame and the width and height of the minimum external frame 
            #  It accelerates the convergence and improves the regression accuracy 
            elif EIoU:
                rho_w2 = ((b2_x2 - b2_x1) - (b1_x2 - b1_x1)) ** beta
                rho_h2 = ((b2_y2 - b2_y1) - (b1_y2 - b1_y1)) ** beta
                cw2 = cw ** beta + eps
                ch2 = ch ** beta + eps
                return iou - (rho2 / c2 + rho_w2 / cw2 + rho_h2 / ch2)
 
            # GIoU https://arxiv.org/pdf/1902.09630.pdf
            c_area = torch.max(cw * ch + eps, union)  # convex area
            return iou - torch.pow((c_area - union) / c_area + eps, alpha)  # GIoU
    else:
        return iou  # torch.log(iou+eps) or iou
 

Last , take utils/loss.py In the document iou=bbox_iou Switch to iou=bbox_alpha_iou that will do .