Yolov5 改进之损失函数 SlideLoss & 注意力机制 MultiSEAM

用于学习记录

---

前言

---

一、SlideLoss

YOLO-FaceV2: A Scale and Occlusion Aware Face Detector

1.1 utils/loss.py

插入以下代码：

import math
class SlideLoss(nn.Module):
    def __init__(self, loss_fcn):
        super(SlideLoss, self).__init__()
        self.loss_fcn = loss_fcn
        self.reduction = loss_fcn.reduction
        self.loss_fcn.reduction = 'none'  # required to apply SL to each element
 
    def forward(self, pred, true, auto_iou=0.5):
        loss = self.loss_fcn(pred, true)
        if auto_iou < 0.2:
            auto_iou = 0.2
        b1 = true <= auto_iou - 0.1
        a1 = 1.0
        b2 = (true > (auto_iou - 0.1)) & (true < auto_iou)
        a2 = math.exp(1.0 - auto_iou)
        b3 = true >= auto_iou
        a3 = torch.exp(-(true - 1.0))
        modulating_weight = a1 * b1 + a2 * b2 + a3 * b3
        loss *= modulating_weight
        if self.reduction == 'mean':
            return loss.mean()
        elif self.reduction == 'sum':
            return loss.sum()
        else:  # 'none'
            return loss

修改 class ComputeLoss

        # Focal loss
        g = h['fl_gamma']  # focal loss gamma
        if g > 0:
            BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)    
   		
   		# 添加以下代码：
        Slide_Loss = h['SlideLoss']
        if Slide_Loss > 0:
            BCEcls, BCEobj = SlideLoss(BCEcls), SlideLoss(BCEobj)

1.2 data/hyps/hyp.scratch-low.yaml

添加以下内容：

SlideLoss: 0.5

二、注意力机制 MultiSEAM

2.1 models/common.py

class Residual(nn.Module):
    def __init__(self, fn):
        super(Residual, self).__init__()
        self.fn = fn
 
    def forward(self, x):
        return self.fn(x) + x
 
 def DcovN(c1, c2, depth, kernel_size=3, patch_size=3):
    dcovn = nn.Sequential(
        nn.Conv2d(c1, c2, kernel_size=patch_size, stride=patch_size),
        nn.SiLU