search

researchmm / SiamDW

[CVPR'19 Oral] Deeper and Wider Siamese Networks for Real-Time Visual Tracking
http://openaccess.thecvf.com/content_CVPR_2019/html/Zhang_Deeper_and_Wider_Siamese_Networks_for_Real-Time_Visual_Tracking_CVPR_2019_paper.html
MIT License
751 stars 180 forks source link

How to improve the tracking? #68

Closed MuhammadAsadJaved closed 4 years ago

MuhammadAsadJaved commented 4 years ago

Hi, I have a question about improving this method. There is a problem that when two objects are crossing each other the tracker lost the object easily and may start tracking the wrong object. For example if we track a person , when another person cross or hide this person , the tracker starts tracking wrong person. We want to use this tracker in end product in real time. Is there any technique or suggestion to over come this problem?

JudasDie commented 4 years ago

Hi, I have a question about improving this method. There is a problem that when two objects are crossing each other the tracker lost the object easily and may start tracking the wrong object. For example if we track a person , when another person cross or hide this person , the tracker starts tracking wrong person. We want to use this tracker in end product in real time. Is there any technique or suggestion to over come this problem?

Hi, thanks for your interest. Occlusion is a difficult problem in tracking. One advice is that you can update template image in tracking. Another advice is that re-identification and global search may be helpful in this case.

MuhammadAsadJaved commented 4 years ago

Ok thank you so much.

On Thu, Jan 2, 2020 at 8:44 PM ZP ZHANG notifications@github.com wrote:

Hi, I have a question about improving this method. There is a problem that when two objects are crossing each other the tracker lost the object easily and may start tracking the wrong object. For example if we track a person , when another person cross or hide this person , the tracker starts tracking wrong person. We want to use this tracker in end product in real time. Is there any technique or suggestion to over come this problem?

Hi, thanks for your interest. Occlusion is a difficult problem in tracking. One advice is that you can update template image in tracking. Another advice is that re-identification and global search may be helpful in this case.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/researchmm/SiamDW/issues/68?email_source=notifications&email_token=AG4GR5ECZ7G4WADK62WLFPLQ3XORBA5CNFSM4KCBZYU2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEH6IJJY#issuecomment-570197159, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG4GR5AJGREP7WNHFGGLSM3Q3XORBANCNFSM4KCBZYUQ .