YOLOv5修改Focal EIOU Loss

修改两处

尝试训练5epoch，loss不收敛，map几乎为0。仅此文章记录一下，Focal EIOU loss代码仍需进一步调参，仅供参考。
参考博客：Focal-EIOU Loss：用于精确边界框回归的高效IOU损失

1. 修改metrics.py

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Model validation metrics
"""

import math
import warnings
from pathlib import Path

import matplotlib.pyplot as plt
import numpy as np
import torch


def fitness(x):
    # Model fitness as a weighted combination of metrics
    w = [0.0, 0.0, 0.1, 0.9]  # weights for [P, R, mAP@0.5, mAP@0.5:0.95]
    return (x[:, :4] * w).sum(1)


def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=()):
    """ Compute the average precision, given the recall and precision curves.
    Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
    # Arguments
        tp:  True positives (nparray, nx1 or nx10).
        conf:  Objectness value from 0-1 (nparray).
        pred_cls:  Predicted object classes (nparray).
        target_cls:  True object classes (nparray).
        plot:  Plot precision-recall curve at mAP@0.5
        save_dir:  Plot save directory
    # Returns
        The average precision as computed in py-faster-rcnn.
    """

    # Sort by objectness
    i = np.argsort(-conf)
    tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]

    # Find unique classes
    unique_classes = np.unique(target_cls)
    nc = unique_classes.shape[0]  # number of classes, number of detections

    # Create Precision-Recall curve and compute AP for each class
    px, py = np.linspace(0, 1, 1000), []  # for plotting
    ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000))
    for ci, c in enumerate(unique_classes):
        i = pred_cls == c
        n_l = (target_cls == c).sum()  # number of labels
        n_p = i.sum()  # number of predictions

        if n_p == 0 or n_l == 0:
            continue
        else:
            # Accumulate FPs and TPs
            fpc = (1 - tp[i]).cumsum(0)
            tpc = tp[i].cumsum(0)

            # Recall
            recall = tpc / (n_l + 1e-16)  # recall curve
            r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0)  # negative x, xp because xp decreases

            # Precision
            precision = tpc / (tpc + fpc)  # precision curve
            p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1)  # p at pr_score

            # AP from recall-precision curve
            for j in range(tp.shape[1]):
                ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
                if plot and j == 0:
                    py.append(np.interp(px, mrec, mpre))  # precision at mAP@0.5

    # Compute F1 (harmonic mean of precision and recall)
    f1 = 2 * p * r / (p + r + 1e-16)
    names = [v for k, v in names.items() if k in unique_classes]  # list: only classes that have data
    names = {i: v for i, v in enumerate(names)}  # to dict
    if plot:
        plot_pr_curve(px, py, ap, Path(save_dir) / 'PR_curve.png', names)
        plot_mc_curve(px, f1, Path(save_dir) / 'F1_curve.png', names, ylabel='F1')
        plot_mc_curve(px, p, Path(save_dir) / 'P_curve.png', names, ylabel='Precision')
        plot_mc_curve(px, r, Path(save_dir) / 'R_curve.png', names, ylabel='Recall')

    i = f1.mean(0).argmax()  # max F1 index
    return p[:, i], r[:, i], ap, f1[:, i], unique_classes.astype('int32')


def compute_ap(recall, precision):
    """ Compute the average precision, given the recall and precision curves
    # Arguments
        recall:    The recall curve (list)
        precision: The precision curve (list)
    # Returns
        Average precision, precision curve, recall curve
    """

    # Append sentinel values to beginning and end
    mrec = np.concatenate(([0.0], recall, [1.0]))
    mpre = np.concatenate(([1.0], precision, [0.0]))

    # Compute the precision envelope
    mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))

    # Integrate area under curve
    method = 'interp'  # methods: 'continuous', 'interp'
    if method == 'interp':
        x = np.linspace(0, 1, 101)  # 101-point interp (COCO)
        ap = np.trapz(np.interp(x, mrec, mpre), x)  # integrate
    else:  # 'continuous'
        i = np.where(mrec[1:] != mrec[:-1])[0]  # points where x axis (recall) changes
        ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])  # area under curve

    return ap, mpre, mrec


class ConfusionMatrix:
    # Updated version of https://github.com/kaanakan/object_detection_confusion_matrix
    def __init__(self, nc, conf=0.25, iou_thres=0.45):
        self.matrix = np.zeros((nc + 1, nc + 1))
        self.nc = nc  # number of classes
        self.conf = conf
        self.iou_thres = iou_thres

    def process_batch(self, detections, labels):
        """
        Return intersection-over-union (Jaccard index) of boxes.
        Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
        Arguments:
            detections (Array[N, 6]), x1, y1, x2, y2, conf, class
            labels (Array[M, 5]), class, x1, y1, x2, y2
        Returns:
            None, updates confusion matrix accordingly
        """
        detections = detections[detections[:, 4] > self.conf]
        gt_classes = labels[:, 0].int()
        detection_classes = detections[:, 5].int()
        iou = box_iou(labels[:, 1:], detections[:, :4])

        x = torch.where(iou > self.iou_thres)
        if x[0].shape[0]:
            matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy()
            if x[0].shape[0] > 1:
                matches = matches[matches[:, 2].argsort()[::-1]]
                matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
                matches = matches[matches[:, 2].argsort()[::-1]]
                matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
        else:
            matches = np.zeros((0, 3))

        n = matches.shape[0] > 0
        m0, m1, _ = matches.transpose().astype(np.int16)
        for i, gc in enumerate(gt_classes):
            j = m0 == i
            if n and sum(j) == 1:
                self.matrix[detection_classes[m1[j]], gc] += 1  # correct
            else:
                self.matrix[self.nc, gc] += 1  # background FP

        if n:
            for i, dc in enumerate(detection_classes):
                if not any(m1 == i):
                    self.matrix[dc, self.nc] += 1  # background FN

    def matrix(self):
        return self.matrix

    def plot(self, normalize=True, save_dir='', names=()):
        try:
            import seaborn as sn

            array = self.matrix / ((self.matrix.sum(0).reshape(1, -1) + 1E-6) if normalize else 1)  # normalize columns
            array[array < 0.005] = np.nan  # don't annotate (would appear as 0.00)

            fig = plt.figure(figsize=(12, 9), tight_layout=True)
            sn.set(font_scale=1.0 if self.nc < 50 else 0.8)  # for label size
            labels = (0 < len(names) < 99) and len(names) == self.nc  # apply names to ticklabels
            with warnings.catch_warnings():
                warnings.simplefilter('ignore')  # suppress empty matrix RuntimeWarning: All-NaN slice encountered
                sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, cmap='Blues', fmt='.2f', square=True,
                           xticklabels=names + ['background FP'] if labels else "auto",
                           yticklabels=names + ['background FN'] if labels else "auto").set_facecolor((1, 1, 1))
            fig.axes[0].set_xlabel('True')
            fig.axes[0].set_ylabel('Predicted')
            fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250)
            plt.close()
        except Exception as e:
            print(f'WARNING: ConfusionMatrix plot failure: {e}')

    def print(self):
        for i in range(self.nc + 1):
            print(' '.join(map(str, self.matrix[i])))


def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=False, EIoU=False, FocalEIoU=False, eps=1e-7):#【改】
    # Returns the 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

    iou = inter / union
    if GIoU or DIoU or CIoU or EIoU or FocalEIoU:#【改】
        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 or FocalEIoU:# 【改】 # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
            c2 = cw ** 2 + ch ** 2 + eps  # convex diagonal squared
            rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 +
                    (b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4  # center distance squared
            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 = v / (v - iou + (1 + eps))
                return iou - (rho2 / c2 + v * alpha)  # CIoU
            elif EIoU or FocalEIoU: #==============【改】EIoU 的实现方式==================
                rho2_w = (w1 - w2) ** 2
                cw2 = cw ** 2 + eps
                rho2_h = (h1 -  h2) ** 2
                ch2 = ch ** 2 + eps

                if EIoU:
                    print('==============【改】EIoU 的实现方式==================')
                    print('DIOU sum()：', (iou - rho2 / c2).sum())
                    print('EIOU sum()：', (iou - (rho2 / c2 + rho2_w / cw2 + rho2_h / ch2)).sum())
                    return iou-(rho2 / c2 + rho2_w/cw2 + rho2_h/ch2)  # EIoU
                else:
                    Loss_EIOU = (1 - iou + (rho2 / c2 + rho2_w / cw2 + rho2_h / ch2)).mean()  # EIoU
                    Loss_FocalEIoU = iou ** 0.5 * Loss_EIOU
                    print('==============【改】FocalEIoU 的实现方式==================')
                    print('DIOU sum()：', (iou - rho2 / c2).sum())
                    print('FocalEIoU sum()：', Loss_FocalEIoU.sum())
                    return Loss_FocalEIoU
        else:  # GIoU https://arxiv.org/pdf/1902.09630.pdf
            c_area = cw * ch + eps  # convex area
            return iou - (c_area - union) / c_area  # GIoU
    else:
        return iou  # IoU


def box_iou(box1, box2):
    # https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
    """
    Return intersection-over-union (Jaccard index) of boxes.
    Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
    Arguments:
        box1 (Tensor[N, 4])
        box2 (Tensor[M, 4])
    Returns:
        iou (Tensor[N, M]): the NxM matrix containing the pairwise
            IoU values for every element in boxes1 and boxes2
    """

    def box_area(box):
        # box = 4xn
        return (box[2] - box[0]) * (box[3] - box[1])

    area1 = box_area(box1.T)
    area2 = box_area(box2.T)

    # inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
    inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
    return inter / (area1[:, None] + area2 - inter)  # iou = inter / (area1 + area2 - inter)


def bbox_ioa(box1, box2, eps=1E-7):
    """ Returns the intersection over box2 area given box1, box2. Boxes are x1y1x2y2
    box1:       np.array of shape(4)
    box2:       np.array of shape(nx4)
    returns:    np.array of shape(n)
    """

    box2 = box2.transpose()

    # Get the coordinates of bounding boxes
    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]

    # Intersection area
    inter_area = (np.minimum(b1_x2, b2_x2) - np.maximum(b1_x1, b2_x1)).clip(0) * \
                 (np.minimum(b1_y2, b2_y2) - np.maximum(b1_y1, b2_y1)).clip(0)

    # box2 area
    box2_area = (b2_x2 - b2_x1) * (b2_y2 - b2_y1) + eps

    # Intersection over box2 area
    return inter_area / box2_area


def wh_iou(wh1, wh2):
    # Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
    wh1 = wh1[:, None]  # [N,1,2]
    wh2 = wh2[None]  # [1,M,2]
    inter = torch.min(wh1, wh2).prod(2)  # [N,M]
    return inter / (wh1.prod(2) + wh2.prod(2) - inter)  # iou = inter / (area1 + area2 - inter)


# Plots ----------------------------------------------------------------------------------------------------------------

def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()):
    # Precision-recall curve
    fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
    py = np.stack(py, axis=1)

    if 0 < len(names) < 21:  # display per-class legend if < 21 classes
        for i, y in enumerate(py.T):
            ax.plot(px, y, linewidth=1, label=f'{names[i]} {ap[i, 0]:.3f}')  # plot(recall, precision)
    else:
        ax.plot(px, py, linewidth=1, color='grey')  # plot(recall, precision)

    ax.plot(px, py.mean(1), linewidth=3, color='blue', label='all classes %.3f mAP@0.5' % ap[:, 0].mean())
    ax.set_xlabel('Recall')
    ax.set_ylabel('Precision')
    ax.set_xlim(0, 1)
    ax.set_ylim(0, 1)
    plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
    fig.savefig(Path(save_dir), dpi=250)
    plt.close()


def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence', ylabel='Metric'):
    # Metric-confidence curve
    fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)

    if 0 < len(names) < 21:  # display per-class legend if < 21 classes
        for i, y in enumerate(py):
            ax.plot(px, y, linewidth=1, label=f'{names[i]}')  # plot(confidence, metric)
    else:
        ax.plot(px, py.T, linewidth=1, color='grey')  # plot(confidence, metric)

    y = py.mean(0)
    ax.plot(px, y, linewidth=3, color='blue', label=f'all classes {y.max():.2f} at {px[y.argmax()]:.3f}')
    ax.set_xlabel(xlabel)
    ax.set_ylabel(ylabel)
    ax.set_xlim(0, 1)
    ax.set_ylim(0, 1)
    plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
    fig.savefig(Path(save_dir), dpi=250)
    plt.close()

2. 修改loss.py

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
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
    return 1.0 - 0.5 * eps, 0.5 * eps


class BCEBlurWithLogitsLoss(nn.Module):
    # BCEwithLogitLoss() with reduced missing label effects.
    def __init__(self, alpha=0.05):
        super().__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
        return loss.mean()


class FocalLoss(nn.Module):
    # Wraps 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().__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().__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


class ComputeLoss:
    # Compute losses
    def __init__(self, model, autobalance=False):
        self.sort_obj_iou = False
        device = next(model.parameters()).device  # get model device
        h = model.hyp  # hyperparameters

        # Define criteria
        BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
        BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))

        # 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
        if g > 0:
            BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)

        det = model.module.model[-1] if is_parallel(model) else model.model[-1]  # Detect() module
        self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, 0.02])  # 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
        device = targets.device
        lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
        tcls, tbox, indices, anchors = self.build_targets(p, targets)  # targets

        # Losses
        for i, pi in enumerate(p):  # layer index, layer predictions
            b, a, gj, gi = indices[i]  # image, anchor, gridy, gridx
            tobj = torch.zeros_like(pi[..., 0], device=device)  # target obj

            n = b.shape[0]  # number of targets
            if n:
                ps = pi[b, a, gj, gi]  # prediction subset corresponding to targets

                # Regression
                pxy = ps[:, :2].sigmoid() * 2 - 0.5
                pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
                pbox = torch.cat((pxy, pwh), 1)  # predicted box
                # iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, EIoU=True)  #【改】# iou(prediction, target)
                # lbox += (1.0 - iou).mean()  # iou loss
                # ==========================【改】 Focal EIOU 实现（注释掉上两行，保存下面两行）==================
                iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, FocalEIoU=True)  #【改】# iou(prediction, target)
                lbox += iou.mean()  # iou loss
                # ==========================【改】 END  ==================

                # 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]
                tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * score_iou  # iou ratio

                # Classification
                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

                # 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)]

            obji = self.BCEobj(pi[..., 4], tobj)
            lobj += obji * self.balance[i]  # obj loss
            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']
        lobj *= self.hyp['obj']
        lcls *= self.hyp['cls']
        bs = tobj.shape[0]  # batch size

        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