Focal Loss笔记

论文：《Focal Loss for Dense Object Detection》

Focal Loss 是何恺明设计的为了解决one-stage目标检测在训练阶段前景类和背景类极度不均衡（如1：1000）的场景的损失函数。它是由二分类交叉熵改造而来的。

标准交叉熵

其中，p是模型预测属于类别y=1的概率。为了方便标记，定义:

交叉熵CE重写为：

α-平衡交叉熵：

有一种解决类别不平衡的方法是引入一个值介于[0; 1]之间的权重因子α：当y=1时，取α; 当y=0时，取1-α。

这种方法，当y=0（即背景类）时，随着α的增大，会对损失进行很大惩罚（降低权重），从而减轻背景类

太多对训练的影响。

类似Pt,可将α-CE重写为：

Focal Loss定义

虽然α-CE起到了平衡正负样本的在损失函数值中的贡献，但是它没办法区分难易样本的样本对损失的贡献。因此就有了Focal Loss，定义如下：

其中，alpha和gamma均为可以调节的超参数。y'为模型预测，其值介于（0-1）之间。

当y=1时，y'->1，表示easy positive，它对权重的贡献->0;

当y=0是，y'->0，表示easy negative，它对权重的贡献->0.

因此，Focal Loss不仅降低了背景类的权重，还降低了easy positive/negative的权重。

gamma是对损失函数的调节，当gamma=0是，Focal Loss与α-CE等价。以下是gamma

对Focal Loss的调节。

Focal Loss的Pytorch实现（蓝色字体）

以下Focal Loss=Focal Loss + Regress Loss;

代码来自：https://github.com/yhenon/pytorch-retinanet

 import numpy as np

 import torch

 import torch.nn as nn

 def calc_iou(a, b):

     area = (b[:, 2] - b[:, 0]) * (b[:, 3] - b[:, 1])

     iw = torch.min(torch.unsqueeze(a[:, 2], dim=1), b[:, 2]) - torch.max(torch.unsqueeze(a[:, 0], 1), b[:, 0])

     ih = torch.min(torch.unsqueeze(a[:, 3], dim=1), b[:, 3]) - torch.max(torch.unsqueeze(a[:, 1], 1), b[:, 1])

     iw = torch.clamp(iw, min=0)

     ih = torch.clamp(ih, min=0)

     ua = torch.unsqueeze((a[:, 2] - a[:, 0]) * (a[:, 3] - a[:, 1]), dim=1) + area - iw * ih

     ua = torch.clamp(ua, min=1e-8)

     intersection = iw * ih

     IoU = intersection / ua

     return IoU

 class FocalLoss(nn.Module):

     #def __init__(self):

     def forward(self, classifications, regressions, anchors, annotations):

         alpha = 0.25

         gamma = 2.0

         batch_size = classifications.shape[0]

         classification_losses = []

         regression_losses = []

         anchor = anchors[0, :, :]

         anchor_widths  = anchor[:, 2] - anchor[:, 0]

         anchor_heights = anchor[:, 3] - anchor[:, 1]

         anchor_ctr_x   = anchor[:, 0] + 0.5 * anchor_widths

         anchor_ctr_y   = anchor[:, 1] + 0.5 * anchor_heights

         for j in range(batch_size):

             classification = classifications[j, :, :]

             regression = regressions[j, :, :]

             bbox_annotation = annotations[j, :, :]

             bbox_annotation = bbox_annotation[bbox_annotation[:, 4] != -1]

             if bbox_annotation.shape[0] == 0:

                 regression_losses.append(torch.tensor(0).float().cuda())

                 classification_losses.append(torch.tensor(0).float().cuda())

                 continue

             classification = torch.clamp(classification, 1e-4, 1.0 - 1e-4)

             IoU = calc_iou(anchors[0, :, :], bbox_annotation[:, :4]) # num_anchors x num_annotations

             IoU_max, IoU_argmax = torch.max(IoU, dim=1) # num_anchors x 1

             #import pdb

             #pdb.set_trace()

             # compute the loss for classification

             targets = torch.ones(classification.shape) * -1

             targets = targets.cuda()

             targets[torch.lt(IoU_max, 0.4), :] = 0

             positive_indices = torch.ge(IoU_max, 0.5)

             num_positive_anchors = positive_indices.sum()

             assigned_annotations = bbox_annotation[IoU_argmax, :]

             targets[positive_indices, :] = 0

             targets[positive_indices, assigned_annotations[positive_indices, 4].long()] = 1

             alpha_factor = torch.ones(targets.shape).cuda() * alpha

             alpha_factor = torch.where(torch.eq(targets, 1.), alpha_factor, 1. - alpha_factor)

 82             focal_weight = torch.where(torch.eq(targets, 1.), 1. - classification, classification)

 83             focal_weight = alpha_factor * torch.pow(focal_weight, gamma)

 84

 85             bce = -(targets * torch.log(classification) + (1.0 - targets) * torch.log(1.0 - classification))

 86

 87             # cls_loss = focal_weight * torch.pow(bce, gamma)

 88             cls_loss = focal_weight * bce

 89

 90             cls_loss = torch.where(torch.ne(targets, -1.0), cls_loss, torch.zeros(cls_loss.shape).cuda())



             classification_losses.append(cls_loss.sum()/torch.clamp(num_positive_anchors.float(), min=1.0))

             # compute the loss for regression

             if positive_indices.sum() > 0:

                 assigned_annotations = assigned_annotations[positive_indices, :]

                 anchor_widths_pi = anchor_widths[positive_indices]

                 anchor_heights_pi = anchor_heights[positive_indices]

                 anchor_ctr_x_pi = anchor_ctr_x[positive_indices]

                 anchor_ctr_y_pi = anchor_ctr_y[positive_indices]

                 gt_widths  = assigned_annotations[:, 2] - assigned_annotations[:, 0]

                 gt_heights = assigned_annotations[:, 3] - assigned_annotations[:, 1]

                 gt_ctr_x   = assigned_annotations[:, 0] + 0.5 * gt_widths

                 gt_ctr_y   = assigned_annotations[:, 1] + 0.5 * gt_heights

                 # clip widths to 1

                 gt_widths  = torch.clamp(gt_widths, min=1)

                 gt_heights = torch.clamp(gt_heights, min=1)

                 targets_dx = (gt_ctr_x - anchor_ctr_x_pi) / anchor_widths_pi

                 targets_dy = (gt_ctr_y - anchor_ctr_y_pi) / anchor_heights_pi

                 targets_dw = torch.log(gt_widths / anchor_widths_pi)

                 targets_dh = torch.log(gt_heights / anchor_heights_pi)

                 targets = torch.stack((targets_dx, targets_dy, targets_dw, targets_dh))

                 targets = targets.t()

                 targets = targets/torch.Tensor([[0.1, 0.1, 0.2, 0.2]]).cuda()

                 negative_indices = 1 - positive_indices

                 regression_diff = torch.abs(targets - regression[positive_indices, :])

                 regression_loss = torch.where(

                     torch.le(regression_diff, 1.0 / 9.0),

                     0.5 * 9.0 * torch.pow(regression_diff, 2),

                     regression_diff - 0.5 / 9.0

                 )

                 regression_losses.append(regression_loss.mean())

             else:

                 regression_losses.append(torch.tensor(0).float().cuda())

 return torch.stack(classification_losses).mean(dim=0, keepdim=True), torch.stack(regression_losses).mean(dim=0, keepdim=True)