Pedestrian-Detection

by xingkongliang

xingkongliang / Pedestrian-Detection

Pedestrian Detection Papers

223 Stars 50 Forks Last release: Not found 13 Commits 0 Releases

Available items

No Items, yet!

The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:

行人检测(Pedestrian Detection)论文整理

@(论文学习记录)[Paper, Pedestrian Detection]

[toc]

相关科研工作者

开放的代码

Paper List

  • [ICCV-2019] Semi-Supervised Pedestrian Instance Synthesis and Detection With Mutual Reinforcement [paper]
  • [ICCV-2019] Weakly Aligned Cross-Modal Learning for Multispectral Pedestrian Detection [paper]
  • [ICCV-2019] Discriminative Feature Transformation for Occluded Pedestrian Detection [paper]
  • [ICCV-2019] Mask-Guided Attention Network for Occluded Pedestrian Detection [paper] [code]
  • [TPAMI-2019] EuroCity Persons: A Novel Benchmark for Person Detection in Traffic Scenes [paper]
  • [CVPR-2019 oral] Adaptive NMS: Refining Pedestrian Detection in a Crowd [paper]
  • [CVPR-2019] High-level Semantic Feature Detection:A New Perspective for Pedestrian Detection [paper] [code]
  • [arxiv] SSA-CNN: Semantic Self-Attention CNN for Pedestrian Detection
  • [CVPR-2019] Pedestrian Detection in Thermal Images using Saliency Maps
  • [TIP-2018] Too Far to See? Not Really:- Pedestrian Detection with Scale-Aware Localization Policy
  • [ECCV-2018] Bi-box Regression for Pedestrian Detection and Occlusion Estimation [code]
  • [ECCV-2018] Learning Efficient Single-stage Pedestrian Detectors by Asymptotic Localization Fitting [code]
  • [ECCV-2018] Graininess-Aware Deep Feature Learning for Pedestrian Detection
  • [ECCV-2018] Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd
  • [ECCV-2018] Small-scale Pedestrian Detection Based on Somatic Topology Localization and Temporal Feature Aggregation
  • [CVPR-2018] Improving Occlusion and Hard Negative Handling for Single-Stage Pedestrian Detectors
  • [CVPR-2018] Occluded Pedestrian Detection Through Guided Attention in CNNs
  • [CVPR-2018] Repulsion Loss: Detecting Pedestrians in a Crowd [code]
  • [TCSVT-2018] Pushing the Limits of Deep CNNs for Pedestrian Detection
  • [Trans Multimedia-2018] Scale-aware Fast R-CNN for Pedestrian Detection
  • [TPAMI-2017] Jointly Learning Deep Features, Deformable Parts, Occlusion and Classification for Pedestrian Detection
  • [BMVC-2017] PCN: Part and Context Information for Pedestrian Detection with CNNs
  • [CVPR-2017] CityPersons: A Diverse Dataset for Pedestrian Detection
  • [CVPR-2017] Learning Cross-Modal Deep Representations for Robust Pedestrian Detection
  • [CVPR-2017] What Can Help Pedestrian Detection?
  • [ICCV-2017] Multi-label Learning of Part Detectors for Heavily Occluded Pedestrian Detection
  • [ICCV-2017] Illuminating Pedestrians via Simultaneous Detection & Segmentation [code]
  • [TPAMI-2017] Towards Reaching Human Performance in Pedestrian Detection
  • [Transactions on Multimedia-2017] Scale-Aware Fast R-CNN for Pedestrian Detection
  • [CVPR-2016] Semantic Channels for Fast Pedestrian Detection
  • [CVPR-2016] How Far are We from Solving Pedestrian Detection?
  • ![CVPR-2016] Pedestrian Detection Inspired by Appearance Constancy and Shape Symmetry
  • ![CVPR-2016] Semantic Channels for Fast Pedestrian Detection
  • ![ECCV-2016] Is Faster R-CNN Doing Well for Pedestrian Detection? [code]
  • [CVPR-2015] Taking a Deeper Look at Pedestrians
  • ![ICCV-2015] Learning Complexity-Aware Cascades for Deep Pedestrian Detection
  • [ICCV-2015] Deep Learning Strong Parts for Pedestrian Detection
  • ![ECCV-2014] Deep Learning of Scene-specific Classifier for Pedestrian Detection
  • [CVPR-2013] Joint Deep Learning for Pedestrian Detection
  • [CVPR-2012] A Discriminative Deep Model for Pedestrian Detection with Occlusion Handling
  • [CVPR-2010] Multi-Cue Pedestrian Classification With Partial Occlusion Handling
  • [CVPR-2009] Pedestrian detection: A benchmark
  • [CVPR-2008] People-Tracking-by-Detection and People-Detection-by-Tracking
  • [ECCV-2006] Human Detection Using Oriented Histograms of Flow and Appearance
  • [CVPR-2005] Histograms of Oriented Gradients for Human Detection

行人检测开源代码

论文

[CVPR-2019] Adaptive NMS: Refining Pedestrian Detection in a Crowd

CVPR19_CSP_Adaptive_NMS - paper: https://arxiv.org/abs/1904.02948

[CVPR-2019] High-level Semantic Feature Detection:A New Perspective for Pedestrian Detection

Alt text - paper: https://arxiv.org/abs/1904.02948 - github: https://github.com/liuwei16/CSP

[CVPR-2019] SSA-CNN: Semantic Self-Attention CNN for Pedestrian Detection

Alt text - paper: https://arxiv.org/abs/1902.09080v1

[CVPR-2019] Pedestrian Detection in Thermal Images using Saliency Maps

  • paper: https://arxiv.org/abs/1904.06859

[TIP-2018] Too Far to See? Not Really:- Pedestrian Detection with Scale-Aware Localization Policy

Alt text| left | 300x0

  • arxiv: https://arxiv.org/abs/1709.00235
  • paper: https://ieeexplore.ieee.org/document/8328854/
  • project website:
  • slides:
  • github caffe:

[Transactions on Multimedia-2017] Scale-Aware Fast R-CNN for Pedestrian Detection

Alt text| left | 300x0

  • arxiv: https://arxiv.org/abs/1510.08160
  • paper: https://ieeexplore.ieee.org/abstract/document/8060595/
  • project website:
  • slides:
  • github caffe:

[ECCV-2018] Bi-box Regression for Pedestrian Detection and Occlusion Estimation

Alt text| left | 300x0 Alt text| left | 300x0

  • arxiv:
  • paper:http://openaccess.thecvf.com/contentECCV2018/papers/CHUNLUANZHOUBi-boxRegressionforECCV2018_paper.pdf
  • slides:
  • github:

[ECCV-2018] Learning Efficient Single-stage Pedestrian Detectors by Asymptotic Localization Fitting

Alt text| left | 300x0

  • arxiv:
  • paper:http://openaccess.thecvf.com/contentECCV2018/papers/WeiLiuLearningEfficientSingle-stageECCV2018_paper.pdf
  • project website:
  • slides:
  • github:

[ECCV-2018] Graininess-Aware Deep Feature Learning for Pedestrian Detection

Alt text| left | 300x0

  • arxiv:
  • paper:http://openaccess.thecvf.com/contentECCV2018/papers/ChunzeLinGraininess-AwareDeepFeatureECCV2018_paper.pdf
  • project website:
  • slides:
  • github:

[ECCV-2018] Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd

20180723-OR-CNN

  • arxiv: http://arxiv.org/abs/1807.08407

[ECCV-2018] Small-scale Pedestrian Detection Based on Somatic Topology Localization and Temporal Feature Aggregation

Alt text| left | 300x0

  • arxiv:https://arxiv.org/abs/1807.01438
  • project website:
  • slides:
  • github caffe:

[CVPR-2018] Improving Occlusion and Hard Negative Handling for Single-Stage Pedestrian Detectors

Alt text| left | 300x0

  • arxiv:
  • paper: http://vision.snu.ac.kr/projects/partgridnet/data/noh_2018.pdf
  • project website: http://vision.snu.ac.kr/projects/partgridnet/
  • slides:
  • github caffe:

[CVPR-2018] Occluded Pedestrian Detection Through Guided Attention in CNNs

Alt text| left | 300x0

  • arxiv:
  • paper: http://openaccess.thecvf.com/contentcvpr2018/papers/ZhangOccludedPedestrianDetectionCVPR2018paper.pdf
  • project website:
  • slides:
  • github caffe:

[CVPR-2018] Repulsion Loss: Detecting Pedestrians in a Crowd

Alt text| left | 300x0

  • arxiv:http://arxiv.org/abs/1711.07752
  • project website:
  • slides:
  • github caffe:

[TPAMI-2017] Jointly Learning Deep Features, Deformable Parts, Occlusion and Classification for Pedestrian Detection

Alt text| left | 300x0

  • paper: https://ieeexplore.ieee.org/abstract/document/8008790/
  • project website:
  • slides:
  • github caffe:

[BMVC-2017] PCN: Part and Context Information for Pedestrian Detection with CNNs

Alt text| left | 300x0

  • arxiv: https://arxiv.org/abs/1804.044838
  • project website:
  • slides:
  • github caffe:

[CVPR-2017] CityPersons: A Diverse Dataset for Pedestrian Detection

Alt text| left | 300x0

  • arxiv: http://arxiv.org/abs/1702.05693
  • project website:
  • slides:
  • github caffe:


[CVPR-2017] Learning Cross-Modal Deep Representations for Robust Pedestrian Detection

Alt text| left | 300x0

  • arxiv: https://arxiv.org/abs/1704.02431
  • project website:
  • slides:
  • github caffe:

Alt text

Alt text

[CVPR-2017] What Can Help Pedestrian Detection?

  • arxiv: https://arxiv.org/abs/1704.02431
  • project website:
  • slides:
  • github caffe:

[TPAMI-2017] Towards Reaching Human Performance in Pedestrian Detection

  • paper: http://ieeexplore.ieee.org/document/7917260/
  • arxiv:
  • project website:
  • slides:
  • github caffe:

[ICCV-2017] Multi-label Learning of Part Detectors for Heavily Occluded Pedestrian Detection

  • paper: http://openaccess.thecvf.com/contentICCV2017/papers/ZhouMulti-LabelLearningofICCV2017paper.pdf
  • arxiv:
  • project website:
  • slides:

[ICCV-2017]Illuminating Pedestrians via Simultaneous Detection & Segmentation

Alt text| left | 300x0

  • arxiv: https://arxiv.org/abs/1706.08564
  • project website: http://cvlab.cse.msu.edu/project-pedestrian-detection.html
  • slides:
  • github caffe: https://github.com/garrickbrazil/SDS-RCNN

[CVPR-2016] Semantic Channels for Fast Pedestrian Detection

Alt text| left | 300x0

  • paper: https://www.cv-foundation.org/openaccess/contentcvpr2016/papers/CosteaSemanticChannelsforCVPR2016paper.pdf
  • project website:
  • slides:
  • github caffe:

[CVPR-2016] Pedestrian Detection Inspired by Appearance Constancy and Shape Symmetry

  • paper: http://openaccess.thecvf.com/contentcvpr2016/papers/CaoPedestrianDetectionInspiredCVPR2016paper.pdf
  • project website:
  • slides:

[CVPR-2016] Semantic Channels for Fast Pedestrian Detection

  • paper: http://openaccess.thecvf.com/contentcvpr2016/papers/CosteaSemanticChannelsforCVPR2016paper.pdf
  • project website:
  • slides:

[ECCV-2016] Is Faster R-CNN Doing Well for Pedestrian Detection?

  • paper:
  • project website:
  • slides:

[CVPR-2016] How Far are We from Solving Pedestrian Detection?

  • paper: https://www.cv-foundation.org/openaccess/contentcvpr2016/app/S06-29.pdf
  • project website:
  • slides:
  • github caffe:

[ICCV-2015] Learning Complexity-Aware Cascades for Deep Pedestrian Detection

  • paper: http://openaccess.thecvf.com/contenticcv2015/papers/CaiLearningComplexity-AwareCascadesICCV2015paper.pdf
  • project website:
  • slides:

[ICCV-2015] Deep Learning Strong Parts for Pedestrian Detection

Alt text| left | 300x0

  • paper: https://www.cv-foundation.org/openaccess/contenticcv2015/html/TianDeepLearningStrongICCV2015paper.htmler.html
  • project website:
  • slides:
  • github caffe:

[CVPR-2013] Joint Deep Learning for Pedestrian Detection Wanli

Alt text| left | 300x0

  • paper: https://www.cv-foundation.org/openaccess/contenticcv2013/html/OuyangJointDeepLearning2013ICCVpaper.html
  • project website:
  • slides:
  • github caffe:

[CVPR-2012] A Discriminative Deep Model for Pedestrian Detection with Occlusion Handling

Alt text| left | 300x0

  • paper: http://mmlab.ie.cuhk.edu.hk/pdf/ouyangWcvpr2012.pdf
  • paper: https://ieeexplore.ieee.org/abstract/document/6248062/
  • project website:
  • slides:
  • github caffe:

[CVPR-2010] Multi-Cue Pedestrian Classification With Partial Occlusion Handling

Alt text| left | 300x0

  • paper: https://ieeexplore.ieee.org/abstract/document/5540111/
  • project website:
  • slides:
  • github caffe:

行人检测数据集

CityPersons

Alt text

CityPersons数据集是在Cityscapes数据集基础上建立的,使用了Cityscapes数据集的数据,对一些类别进行了精确的标注。该数据集是在[CVPR-2017] CityPersons: A Diverse Dataset for Pedestrian Detection这篇论文中提出的,更多细节可以通过阅读该论文了解。

上图中左侧是行人标注,右侧是原始的CityScapes数据集。

#评测文件
$/Cityscapes/shanshanzhang-citypersons/evaluation/eval_script/coco.py
$/Cityscapes/shanshanzhang-citypersons/evaluation/eval_script/eval_demo.py
$/Cityscapes/shanshanzhang-citypersons/evaluation/eval_script/eval_MR_multisetup.py

#注释文件 $/Cityscapes/shanshanzhang-citypersons/annotations $/Cityscapes/shanshanzhang-citypersons/annotations/anno_train.mat $/Cityscapes/shanshanzhang-citypersons/annotations/anno_val.mat $/Cityscapes/shanshanzhang-citypersons/annotations/README.txt #图片数据

$/Cityscapes/leftImg8bit/train/* $/Cityscapes/leftImg8bit/val/* $/Cityscapes/leftImg8bit/test/*

注释文件格式 ``` CityPersons annotations (1) data structure: one image per cell in each cell, there are three fields: cityname; imname; bbs (bounding box annotations)

(2) bounding box annotation format:    one object instance per row:    [classlabel, x1,y1,w,h, instanceid, x1vis, y1vis, wvis, hvis]

(3) class label definition:   classlabel =0: ignore regions (fake humans, e.g. people on posters, reflections etc.) classlabel =1: pedestrians classlabel =2: riders classlabel =3: sitting persons classlabel =4: other persons with unusual postures classlabel =5: group of people

(4) boxes:   visible boxes [x1vis, y1vis, wvis, hvis] are automatically generated from segmentation masks; (x1,y1) is the upper left corner. if classlabel==1 or 2 [x1,y1,w,h] is a well-aligned bounding box to the full body ; else [x1,y1,w,h] = [x1vis, y1vis, wvis, h_vis]; ```

Caltech

Alt text

KITTI

Alt text

Alt text

EuroCity

EuroCity 官网

EuroCity Paper

  • [TPAMI-2019] EuroCity Persons: A Novel Benchmark for Person Detection in Traffic Scenes

With over 238200 person instances manually labeled in over 47300 images, EuroCity Persons is nearly one order of magnitude larger than person datasets used previously for benchmarking. Diversity is gained by recording this dataset throughout Europe.

EuroCity-01

EuroCity-02

| Object Class | # objects (day) | # objects (night) | # objects (sum) | |:------------:|:---------------:|:-----------------:|:---------------:| | Pedestrian | 183004 | 35309 | 218313 | | Rider | 18216 | 1564 | 19780 |

CrowdHuman

CrowdHuman 主页

CrowdHuman Paper

CrowdHuman-20190918-01

CrowdHuman-20190918-02

CrowdHuman-20190918-03

性能比较

数据来自 CityPersons 官网。

| Method | MR (Reasonable) | MR (Reasonablesmall) | MR (Reasonableocc=heavy) | MR (All) | |:------------------:|:---------------:|:---------------------:|:-------------------------:|:--------:| | YT-PedDet | 8.41% | 10.60% | 37.88% | 37.22% | | STNet | 9.78% | 10.95% | 36.16% | 31.36% | | DVRNet | 10.99% | 15.68% | 43.77% | 41.48% | | HBA-RCNN | 11.06% | 14.77% | 43.61% | 39.54% | | OR-CNN | 11.32% | 14.19% | 51.43% | 40.19% | | Repultion Loss | 11.48% | 15.67% | 52.59% | 39.17% | | Adapted FasterRCNN | 12.97% | 37.24% | 50.47% | 43.86% | | MS-CNN | 13.32% | 15.86% | 51.88% | 39.94% |

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.