本文主要总结2020年来的一些较好地目标跟踪算法的效果,最后总结完,会争取上传一个表格!
最后更新时间:20201103
1、Probabilistic Regression for Visual Tracking(PrDiMP)[CVPR2020]
文章链接:https://openaccess.thecvf.com/content_CVPR_2020/html/Danelljan_Probabilistic_Regression_for_Visual_Tracking_CVPR_2020_paper.html
code链接:https://github.com/visionml/pytracking
ResNet18版本跑大约40FPS,ResNet50版本跑大约30FPS。
LaSOT和UAV123:
TrackingNet:
VOT2018:
GOT10k:
OTB-100和NFS:
2、Know Your Surroundings:Exploiting Scene Information for Object Tracking(KYS)[ECCV2020]
文章链接:https://arxiv.org/abs/2003.11014
code链接:https://github.com/visionml/pytracking
在Nvidia RTX 2080 GPU上跑20FPS。
VOT2018:
GOT10k:
TrackingNet:
GOT10k、OTB-100和NFS:
3、D3S—A Discriminative Single Shot Segmentation Tracker [CVPR2020]
文章链接:https://openaccess.thecvf.com/content_CVPR_2020/html/Lukezic_D3S_-_A_Discriminative_Single_Shot_Segmentation_Tracker_CVPR_2020_paper.html
code链接:https://github.com/alanlukezic/d3s
在一张Nvidia GTX 1080 GPU上跑25fps,需要1.3s载入网络和初始化。
VOT2016:
VOT2018:
GOT10k:
TrackingNet:
4、Siamese Box Adaptive Network for Visual Tracking(SiamBAN)[CVPR2020]
文章链接:https://openaccess.thecvf.com/content_CVPR_2020/html/Chen_Siamese_Box_Adaptive_Network_for_Visual_Tracking_CVPR_2020_paper.html
code链接:https://github.com/hqucv/siamban
跑40FPS。
VOT2018:
VOT2019 real-time:
OTB100:
NFS:
UAV123:
LaSOT:
5、SiamCAR:Simese Fully Convolutional Classification and Regression for Visual Tracking [CVPR2020]
文章链接:https://openaccess.thecvf.com/content_CVPR_2020/html/Guo_SiamCAR_Siamese_Fully_Convolutional_Classification_and_Regression_for_Visual_Tracking_CVPR_2020_paper.html
code链接: https://github.com/ohhhyeahhh/SiamCAR
GOT10k:
LaSOT:
OTB-50:
UAV123:
6、 Siam R-CNN: Visual Tracking by Re-Detection [CVPR2020]
文章链接:https://openaccess.thecvf.com/content_CVPR_2020/html/Voigtlaender_Siam_R-CNN_Visual_Tracking_by_Re-Detection_CVPR_2020_paper.html
code链接:www.vision.rwth-aachen.de/page/siamrcnn
OTB100:
UAV123:
NFS:
TrackingNet:
GOT-10k:
VOT2018:
LaSOT:
7、Ocean: Object-aware Anchor-free Tracking [ECCV2020]
文章链接:
code链接:
网好慢,论文还没有下载下来。。。
8、Deformable Siamese Attention Networks for Visual Object Tracking(SiamAttn)[CVPR2020]
文章链接:https://openaccess.thecvf.com/content_CVPR_2020/papers/Yu_Deformable_Siamese_Attention_Networks_for_Visual_Object_Tracking_CVPR_2020_paper.pdf
code链接:
在OTB100、UAV、LaSOT和TrackingNet上,跑45FPS;在VOT上跑33FPS。
OTB-100:
VOT2016和VOT2018:
UAV123:
LaSOT:
TrackingNet: