yolo增加slide loss，改善样本不平衡问题
slide loss的主要作用是让模型更加关注难例，可以轻微的改善模型在难例检测上的效果
论文地址：https://arxiv.org/pdf/2208.02019.pdf
代码：GitHub - Krasjet-Yu/YOLO-FaceV2: YOLO-FaceV2: A Scale and Occlusion Aware Face Detector
样本不平衡问题，即在大多数情况下，容易样本的数量很大，而困难样本相对稀疏，引起了很多关注。在本文的工作中，设计了一个看起来像“slide”的Slide Loss函数来解决这个问题。简单样本和困难样本之间的区别是基于预测框和ground truth 框的IoU大小。为了减少超参数，将所有边界框的 IoU 值的平均值作为阈值 µ，小于µ的取负样本，大于µ的取正样本。
然而，由于分类不明确，边界附近的样本往往会遭受较大的损失。希望模型能够学习优化这些样本，并更充分地使用这些样本来训练网络。然而，此类样本的数量相对较少。因此，尝试为困难样本分配更高的权重。首先通过参数μ将样本分为正样本和负样本。然后，通过加权函数Slide对边界处的样本进行强调，如图 4 所示。Slide加权函数可以表示为公式5。
在utils/loss.py增加
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在data\hyps\hyp.scratch-low.yaml中增加
slide_ratio: 1 # >=1启用slide loss, <1关闭在utils/loss.py的ComputeLoss函数中做如下修改：
class ComputeLoss:
# Compute losses
def __init__(self, model, autobalance=False):
super(ComputeLoss, self).__init__()
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
# slide loss
self.slide_ratio = h['slide_ratio']
if self.slide_ratio > 0:
BCEcls, BCEobj = SlideLoss(BCEcls), SlideLoss(BCEobj)
# 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, .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, model.gr, 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)
lrepBox, lrepGT = 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, CIoU=True) # iou(prediction, target)
auto_iou = iou.mean()
lbox += (1.0 - iou).mean() # iou loss
# Objectness
tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype) # 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
if self.slide_ratio > 0:
lcls += self.BCEcls(ps[:, 5:], t, auto_iou) # BCE
else:
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)]
if self.slide_ratio > 0 and n:
obji = self.BCEobj(pi[..., 4], tobj, auto_iou)
else:
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
loss = lbox + lobj + lcls
return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach()主要修改如下：
1、__init__中增加
# slide loss
self.slide_ratio = h['slide_ratio']
if self.slide_ratio > 0:
BCEcls, BCEobj = SlideLoss(BCEcls), SlideLoss(BCEobj)2、计算完iou后增加
auto_iou = iou.mean()3、在类别损失函数上
if self.slide_ratio > 0:
lcls += self.BCEcls(ps[:, 5:], t, auto_iou) # BCE
else:
lcls += self.BCEcls(ps[:, 5:], t) # BCE4、前背景损失函数上
if self.slide_ratio > 0 and n:
obji = self.BCEobj(pi[..., 4], tobj, auto_iou)
else:
obji = self.BCEobj(pi[..., 4], tobj)