FCOS: Fully Convolutional One-Stage Object Detection

原文地址:https://arxiv.org/abs/1904.01355

github地址:https://github.com/tianzhi0549/FCOS/

Research background:

基于锚节点的目标检测有如下四个缺点:

1.detection performance is sensitive to the sizes, aspect ratios and number of anchor boxes. For example, in RetinaNet, varying these hyper-parameters affects the performance up to 4% in AP on the COCO benchmark . As a result, these hyper-parameters need to be carefully tuned in anchor-based detectors.

2.Even with careful design, because the scales and aspect ratios of anchor boxes are kept fixed, detectors encounter difficulties to deal with object candidates with large shape variations, particularly for small objects. The pre-defined anchor boxes also hamper the generalization ability of detectors, as they need to be redesigned on new detection tasks with different object sizes or aspect ratios.

3. In order to achieve a high recall rate, an anchor-based detector is required to densely place anchor boxes on the input image (e.g., more than 180K anchor boxes in feature pyramid networks (FPN) [11] for an image with its shorter side being 800). Most of these anchor boxes are labelled as negative samples during training. The excessive number of negative samples aggravates the imbalance between positive and negative samples in training.

4. An excessively large number of anchor boxes also significantly increase the amount of computation and memory footprint when computing the intersection-over-union (IOU) scores between all anchor boxes and ground-truth boxes during training.

简而言之就是:1.对锚窗口大小和比率敏感,需要仔细微调才能出好结果;2.需要仔细设计,在新数据集上需要重新设计锚节点。3.正负样本不平衡。4.锚节点过多导致浪费显存。

锚节点是什么?为什么需要锚节点呢?

anchor的提出每个像素点(特征像素点)的不同大小的候选窗口,通过使用anchor方法,可以在模型中提取适应不同的目标大小候选窗口,在不同anchor-based的论文中,anchor设定是不固定的。

既然只是提出候选窗口区域,有没有方法可以去除他呢?

本文就是以为anchor-based的缺点,而提出无anchor的检测方法。

FullyConvolutionalOne-StageObjectDetector

不同于anchor-based object detection预测x,y,w,h,本文采用了预测l,t,r,b.分别代表该像素点到检测窗口四个边界的距离。公式如下:

因此,损失计算就变为:

tip:如果细心想要复现的人就会发现,这个真值标签在训练过程中要怎么制作,因为该特征点是在卷积和下采样过程中,位置发生了变化。这时候可以通过如下公式将该特征点位置映射回原始图像中的(x,y):

S为缩小比率(下采样后的相当于原图大小的比率)

 Multi-level Prediction with FPN for FCOS

仅仅依靠上面 的方法是不够的,不能代替anchor的作用,因此本文使用了通过FPN的多层次预测方法来计算不同大小特征图的预测窗口。这样不仅能够检测出小窗口目标,也能检测出大窗口目标。但是如果每个特征图都直接预测出所有大小的检测窗口,会出现一定的重叠,因此,每个特征图只预测出相应大小比例的目标检测窗口,通过如下方法限制检测窗口大小:

m的取值为

 

这时候可以测试出本文方法的实验结果:

Center-ness for FCOS

从上面方法可以看到,本文方法得到了很大的提升,但是准确率却是不够的,因此作者仔细查看预测结果发现了

We observed that it is due to a lot of low-quality predicted bounding boxes produced by locations far away from the center of an object. 预测过程中出现了许多远离目标中心的低质量检测窗口。

因此,作者提出center-ness target,专门提出一个预测通道来预测center-ness值。

 

真值center-ness的计算公式如下:

通过缩小预测的centerness来抑制原理中心的检测窗口。而预测离中心的窗口多余窗口会通过nms(极大值抑制)来吸收。

结果:

本文还做了其他实验验证本文方法的有效性,这里就不多说了(ps:懒得写了,哈哈哈)

完(笑)

 

 

 

 

 

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值