search

iyybpatrick / M-OBJ-TRK-ALGO

source code for a Multiple Object Tracking Algorithm
13 stars 6 forks source link

Add some description #2

Open feymanpriv opened 7 years ago

feymanpriv commented 7 years ago

Thank you for providing us with such a great C++ code!However, as a novice in the tracking field, I have met a lot of difficulties to learn your code.Would you please add some description such as a detailed README to help us to use this code and learn the algorithm?

iyybpatrick commented 7 years ago

Hi there,

All you need can be found in this paper. https://arxiv.org/abs/1611.02364

Thanks,

Yuebin

On 2017年8月17日 at 上午4:39:11, feyman (notifications@github.com) wrote:

Thank you for providing us with such a great C++ code!However, as a novice in the tracking field, I have met a lot of difficulties to learn your code.Would you please add some description such as a detailed README to help us to use this code and learn the algorithm?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/iyybpatrick/MKCF/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AOhjgt01ilI9EQ201WpcB3KrfDJpMJwRks5sY_wugaJpZM4O58n9 .

msnopko commented 7 years ago

Hello, is this algorithm able to work with real time video stream?

iyybpatrick commented 7 years ago

Yes, of course.

On 2017年8月17日 at 上午9:03:14, M1R3C (notifications@github.com) wrote:

Hello, is this algorithm able to work with real time video stream?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/iyybpatrick/MKCF/issues/2#issuecomment-323066881, or mute the thread https://github.com/notifications/unsubscribe-auth/AOhjgk_KURpyJhGh7GaDeiNzOnw9wZD-ks5sZDnygaJpZM4O58n9 .

msnopko commented 7 years ago

But in line 634 you are reading an image from your disk and I am unable to create those images in advance. For example if we have 30 fps video stream, we need to create 30 foreground images and analyze them with trackers in one second. What is the best way to create those "foreground" images during program execution? I've read about Lobster in your papers, but I don't know how it actually works.

iyybpatrick commented 7 years ago

Hi again,

Actually I downloaded the generated background images on this website. https://www.jpjodoin.com/urbantracker/dataset.html I guess you may ask the author for detailed methodology if you want to test the algorithm on other videos. This work is beyond my paper, I’m sorry I cannot help you.

Yuebin

On 2017年8月17日 at 上午9:46:59, M1R3C (notifications@github.com) wrote:

But in line 634 you are reading an image from your disk and I am unable to create those images in advance. For example if we have 30 fps video stream, we need to create 30 foreground images and analyze them with trackers in one second. What is the best way to create those "foreground" images during program execution? I've read about Lobster in your papers, but I don't know how it actually works.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/iyybpatrick/MKCF/issues/2#issuecomment-323078344, or mute the thread https://github.com/notifications/unsubscribe-auth/AOhjgme75x8IyCDyGeDIqoExd2uE69acks5sZERLgaJpZM4O58n9 .

feymanpriv commented 7 years ago

I want to finish the same work with you(M1R3C).And our videos are much longer than the ones in the dataset.It is not so wise to get per fg frame mask in advance and then apply the algorithm. But if we use background subtraction, analyze the fgs and assign trackers per frame at the meantime, it is difficult to achieve the real-time standard.