How Far are We from Solving Pedestrian Detection?
2016-08-04 15:25
615 查看
CVPR 2016 我们离解决行人检测问题到底还有多远?
How Far are We from Solving Pedestrian Detection?
项目网页:https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/people-detection-pose-estimation-and-tracking/how-far-are-we-from-solving-pedestrian-detection/
Matlab code: https://bitbucket.org/shanshanzhang/code_filteredchannelfeatures
行人检测问题一直是一个比较热门的研究话题,行人检测最近几年进步比较大,那么还有多大的提升空间了?最近建立了一个 human baseline(人自己去看,来框出行人),发现大约目前最先进的算法与人工检测差 10%。
于是我们分析了一下目前算法主要存在的问题,找出改进的空间:1) high confidence false positives 一个主要因素是 localisation ,这里我们通过改进 training set alignment quality 来解决;2) 背景的干扰,通过深入分析CNN网络来改进
这里我们首先建立了一个 human baseline, 这是我们的终极目标,达到或者超过人工检测率。
3.2. Failure analysis
紧接着我们分析了一下 当前最好的算法检测失败的原因
Conclusion: For most top performing methods localisation and background-vs-foreground errors have equal impact on the detection quality. They are equally important.
3.3. Improved Caltech-USA annotations
原来的数据库真值有不太准确的地方,我们对此进行了改进。
4.1. Impact of training annotations
4.2. Convnets for pedestrian detection
Conclusion :CNN网络在图像分类和广义目标检测问题显示出很强的能力,但是针对小目标检测定位问题表现的不是很好,加入了Bounding box regression (and
NMS)有所改善,但是 背景的干扰仍然是检测失败主要的原因。
How Far are We from Solving Pedestrian Detection?
项目网页:https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/people-detection-pose-estimation-and-tracking/how-far-are-we-from-solving-pedestrian-detection/
Matlab code: https://bitbucket.org/shanshanzhang/code_filteredchannelfeatures
行人检测问题一直是一个比较热门的研究话题,行人检测最近几年进步比较大,那么还有多大的提升空间了?最近建立了一个 human baseline(人自己去看,来框出行人),发现大约目前最先进的算法与人工检测差 10%。
于是我们分析了一下目前算法主要存在的问题,找出改进的空间:1) high confidence false positives 一个主要因素是 localisation ,这里我们通过改进 training set alignment quality 来解决;2) 背景的干扰,通过深入分析CNN网络来改进
这里我们首先建立了一个 human baseline, 这是我们的终极目标,达到或者超过人工检测率。
3.2. Failure analysis
紧接着我们分析了一下 当前最好的算法检测失败的原因
Conclusion: For most top performing methods localisation and background-vs-foreground errors have equal impact on the detection quality. They are equally important.
3.3. Improved Caltech-USA annotations
原来的数据库真值有不太准确的地方,我们对此进行了改进。
4.1. Impact of training annotations
4.2. Convnets for pedestrian detection
Conclusion :CNN网络在图像分类和广义目标检测问题显示出很强的能力,但是针对小目标检测定位问题表现的不是很好,加入了Bounding box regression (and
NMS)有所改善,但是 背景的干扰仍然是检测失败主要的原因。
相关文章推荐
- How Far are We from Solving Pedestrian Detection?
- How Far are We from Solving Pedestrian Detection?
- 行人检测论文笔记:How Far are We from Solving Pedestrian Detection?
- How Far are We from Solving Pedestrian Detection?
- How Far are We from Solving Pedestrian Detection?
- paper 157:文章解读--How far are we from solving the 2D & 3D Face Alignment problem?-(and a dataset of 230,000 3D facial landmarks)
- 人脸对齐--How far are we from solving the 2D & 3D Face Alignment problem
- 计算机视觉和人工智能的状态:我们已经走得很远了 The state of Computer Vision and AI: we are really, really far away.
- Ten Years of Pedestrian Detection, What Have We Learned?
- How to avoid Excel from prompting "The file you are trying to open, , is in a different format than specified by the file extension"? (转)
- How can we use transfer function to skip search page while going from one component to another?
- [转] How Poor We Really Are
- In how many ways can we select some of these coins so that they are X yen in total?
- Ten Years of Pedestrian Detection,What Have We Learned? 知识笔记
- you and me ,from one world, we are family
- 行人检测论文笔记:Ten Years of Pedestrian Detection, What Have We Learned?
- 论文笔记:Ten years of pedestrian detection, what have we learned?
- How good are detection proposals, really?
- [BetterExplained]How can we add the numbers from 1 to 100?
- How Can We Benefit From 24-Port Managed PoE Switch?