TensorRT_Transform/utils/loss.py

# Loss functions

import torch
import torch.nn as nn

from utils.metrics import bbox_iou
from utils.torch_utils import is_parallel


def smooth_BCE(eps=0.1):  # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441
    # return positive, negative label smoothing BCE targets 标签平滑，https://wenku.baidu.com/view/27fdf1deadf8941ea76e58fafab069dc51224773.html
    #机器学习样本中少量错误标签，影响预测效果，训练时假设可能存在错误，避免过分相信。如果是交叉熵，可以简单实现，成为标签平滑。
    return 1.0 - 0.5 * eps, 0.5 * eps


class BCEBlurWithLogitsLoss(nn.Module):
    # BCEwithLogitLoss() with reduced missing label effects.
    #BCEWithLogitsLoss这个loss类将sigmoid操作和BCELoss（二进制交叉熵损失）集合到了一个类
    def __init__(self, alpha=0.05):
        super(BCEBlurWithLogitsLoss, self).__init__()
        self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none')  # must be nn.BCEWithLogitsLoss()
        self.alpha = alpha

    def forward(self, pred, true):
        loss = self.loss_fcn(pred, true)
        pred = torch.sigmoid(pred)  # prob from logits
        dx = pred - true  # reduce only missing label effects
        # dx = (pred - true).abs()  # reduce missing label and false label effects
        alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4))
        loss *= alpha_factor  #loss乘以alpha_factor这个系数，  考虑到图片中有目标但是没有做标签的情况，false negative
        return loss.mean()


class FocalLoss(nn.Module):
    # Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
    #FocalLoss主要是为了解决one-stage的目标检测中正负样本比例严重失衡的问题，损失函数降低了大量简单负样本在训练过程中的比例
    def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
        super(FocalLoss, self).__init__()
        self.loss_fcn = loss_fcn  # must be nn.BCEWithLogitsLoss()
        self.gamma = gamma
        self.alpha = alpha
        self.reduction = loss_fcn.reduction
        self.loss_fcn.reduction = 'none'  # required to apply FL to each element

    def forward(self, pred, true):
        loss = self.loss_fcn(pred, true)
        # p_t = torch.exp(-loss)
        # loss *= self.alpha * (1.000001 - p_t) ** self.gamma  # non-zero power for gradient stability

        # TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py
        pred_prob = torch.sigmoid(pred)  # prob from logits
        p_t = true * pred_prob + (1 - true) * (1 - pred_prob)
        alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
        modulating_factor = (1.0 - p_t) ** self.gamma
        loss *= alpha_factor * modulating_factor

        if self.reduction == 'mean':
            return loss.mean()
        elif self.reduction == 'sum':
            return loss.sum()
        else:  # 'none'
            return loss


class QFocalLoss(nn.Module):
    # Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
    def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
        super(QFocalLoss, self).__init__()
        self.loss_fcn = loss_fcn  # must be nn.BCEWithLogitsLoss()
        self.gamma = gamma
        self.alpha = alpha
        self.reduction = loss_fcn.reduction
        self.loss_fcn.reduction = 'none'  # required to apply FL to each element

    def forward(self, pred, true):
        loss = self.loss_fcn(pred, true)

        pred_prob = torch.sigmoid(pred)  # prob from logits
        alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
        modulating_factor = torch.abs(true - pred_prob) ** self.gamma
        loss *= alpha_factor * modulating_factor

        if self.reduction == 'mean':
            return loss.mean()
        elif self.reduction == 'sum':
            return loss.sum()
        else:  # 'none'
            return loss


#计算损失=（分类损失+置信度损失+框坐标回归损失）
class ComputeLoss:
    # Compute losses
    def __init__(self, model, autobalance=False):
        super(ComputeLoss, self).__init__()
        self.sort_obj_iou = False
        device = next(model.parameters()).device  # get model device
        h = model.hyp  # hyperparameters 获得超参数！！！

        # Define criteria 定义类别及目标性得分损失函数
        BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))  #将'cls_pw'这两个参数传进来，在hyp.scratch.yaml里
        BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device)) #将'obj_pw'这两个参数传进来，在hyp.scratch.yaml里

        # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
        self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0))  # positive, negative BCE targets

        # Focal loss
        g = h['fl_gamma']  # focal loss gamma
        #这里g>0才考虑focal loss
        if g > 0:
            BCEcls, BCEobj = QFocalLoss(BCEcls, g), QFocalLoss(BCEobj, g)

        det = model.module.model[-1] if is_parallel(model) else model.model[-1]  # Detect() module
        #设置3个特征图对应的损失函数的损失系数  80x80、40x40、20x20有相应的系数  显然80特征图系数最大
        self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, .02])  # P3-P7

        self.ssi = list(det.stride).index(16) if autobalance else 0  # stride 16 index
        self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance
        for k in 'na', 'nc', 'nl', 'anchors':
            setattr(self, k, getattr(det, k))

    def __call__(self, p, targets):  # predictions, targets, model  __call__可以实例化对象名后直接调用这个函数，格式是：实例化对象名（参数）
        #p是网络的输出，targets是这个batch中所有图片标注的目标框信息
        #获取设备，用的是cuda
        device = targets.device
        #初始化各部分损失
        #类别损失、box回归损失、目标性得分损失（即置信度）
        lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
        #获得标签分类信息、边框信息（不同尺度上的预测框）、索引信息、anchors
        tcls, tbox, indices, anchors = self.build_targets(p, targets)  # targets

        # Losses
        #遍历每一个预测输出
        for i, pi in enumerate(p):  # layer index, layer predictions  在每一层特征图上迭代，比如先80x80，再40x40，最后20x20
            #根据indices获取索引，方便找到对应网格的输出
            b, a, gj, gi = indices[i]  # image, anchor, gridy, gridx
            tobj = torch.zeros_like(pi[..., 0], device=device)  # target obj   tobj初始化为0

            n = b.shape[0]  # number of targets
            if n:
                #找到对应网格的输出，取出对应位置的预测值。若pi里为80x80，则那个维度数据对应此特征图上，下面pxy和pwh进行框微调，有专门公式
                ps = pi[b, a, gj, gi]  # prediction subset corresponding to targets

                # Regression
                #对输出的xywh做反算
                #想计算预测框的xy，这里是微调？
                pxy = ps[:, :2].sigmoid() * 2. - 0.5
                #想计算预测框的wh，这里是微调？ 通过偏移值，求出这个框真正的xywh
                pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
                pbox = torch.cat((pxy, pwh), 1)  # predicted box
                #计算边框损失，注意这个CIoU=True，计算的是是CIoU，bbox_iou可以选择传参呢！。 注意：tbox[i]里面是groundtruth
                iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True)  # iou(prediction, target)
                lbox += (1.0 - iou).mean()  # iou loss   损失函数在这里了，通过iou算出iou的loss。求了mean，就变成一个均值。

                # Objectness
                score_iou = iou.detach().clamp(0).type(tobj.dtype)
                if self.sort_obj_iou:
                    sort_id = torch.argsort(score_iou)
                    b, a, gj, gi, score_iou = b[sort_id], a[sort_id], gj[sort_id], gi[sort_id], score_iou[sort_id]
                #根据model.gr设置objectness的标签值，有目标的conf分支权重。
                #不同anchor和gt bbox匹配度不一样，预测框和gt bbox的匹配度也不一样，如果权重设置一样肯定不是最优的
                #故将预测框和bbox的iou作为权重乘到conf分支，用于表征预测质量
                tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * score_iou  # iou ratio

                # Classification
                #如果类别数>1，才计算分类损失（即多类别损失）
                if self.nc > 1:  # cls loss (only if multiple classes)
                    t = torch.full_like(ps[:, 5:], self.cn, device=device)  # targets
                    t[range(n), tcls[i]] = self.cp
                    lcls += self.BCEcls(ps[:, 5:], t)  # BCE对每个类别单独计算loss

                # Append targets to text file
                # with open('targets.txt', 'a') as file:
                #     [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]

            #计算objectness的损失
            obji = self.BCEobj(pi[..., 4], tobj)
            lobj += obji * self.balance[i]  # obj loss  考虑了balance的值，即不同的特征图大小考虑不同的权重！！！
            if self.autobalance:
                self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item()

        if self.autobalance:
            self.balance = [x / self.balance[self.ssi] for x in self.balance]
        lbox *= self.hyp['box']  #求最后总的损失时，还进行了加权
        lobj *= self.hyp['obj']  #超参里面有设置
        lcls *= self.hyp['cls']  #超参里面有设置
        bs = tobj.shape[0]  # batch size

        #总的loss=lbox + lobj + lcls，再乘以batchsize，返回
        return (lbox + lobj + lcls) * bs, torch.cat((lbox, lobj, lcls)).detach()

    def build_targets(self, p, targets):
        # Build targets for compute_loss(), input targets(image,class,x,y,w,h)
        na, nt = self.na, targets.shape[0]  # number of anchors, targets
        tcls, tbox, indices, anch = [], [], [], []
        gain = torch.ones(7, device=targets.device)  # normalized to gridspace gain
        ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt)  # same as .repeat_interleave(nt)
        targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2)  # append anchor indices

        g = 0.5  # bias
        off = torch.tensor([[0, 0],
                            [1, 0], [0, 1], [-1, 0], [0, -1],  # j,k,l,m
                            # [1, 1], [1, -1], [-1, 1], [-1, -1],  # jk,jm,lk,lm
                            ], device=targets.device).float() * g  # offsets

        for i in range(self.nl):
            anchors = self.anchors[i]
            gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]]  # xyxy gain

            # Match targets to anchors
            t = targets * gain
            if nt:
                # Matches
                r = t[:, :, 4:6] / anchors[:, None]  # wh ratio
                j = torch.max(r, 1. / r).max(2)[0] < self.hyp['anchor_t']  # compare
                # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t']  # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
                t = t[j]  # filter

                # Offsets
                gxy = t[:, 2:4]  # grid xy
                gxi = gain[[2, 3]] - gxy  # inverse
                j, k = ((gxy % 1. < g) & (gxy > 1.)).T
                l, m = ((gxi % 1. < g) & (gxi > 1.)).T
                j = torch.stack((torch.ones_like(j), j, k, l, m))
                t = t.repeat((5, 1, 1))[j]
                offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
            else:
                t = targets[0]
                offsets = 0

            # Define
            b, c = t[:, :2].long().T  # image, class
            gxy = t[:, 2:4]  # grid xy
            gwh = t[:, 4:6]  # grid wh
            gij = (gxy - offsets).long()
            gi, gj = gij.T  # grid xy indices

            # Append
            a = t[:, 6].long()  # anchor indices
            indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1)))  # image, anchor, grid indices
            tbox.append(torch.cat((gxy - gij, gwh), 1))  # box
            anch.append(anchors[a])  # anchors
            tcls.append(c)  # class

        return tcls, tbox, indices, anch