zhengthomastang / 2018AICity_TeamUW

The winning method in Track 1 and Track 3 at the 2nd AI City Challenge Workshop in CVPR 2018 - Official Implementation
http://openaccess.thecvf.com/content_cvpr_2018_workshops/w3/html/Tang_Single-Camera_and_Inter-Camera_CVPR_2018_paper.html
550 stars 198 forks source link

Tracklet clustering #26

Open harshnandan opened 3 years ago

harshnandan commented 3 years ago

I am trying to understand the tracklet clustering algorithm (section 3.5) of your CVPR paper. I have two questions to start with:

image

For sake of simplicity, assume that I am just considering smoothness loss for figuring out if two tracklets in their individual set can be merged together.

  1. Is it right to say that if each tracklet is its own set (let's say that is where I start my optimization) the loss of each set will be zero? I am trying to see what is the value of 2nd term in equation 18 of the paper.

  2. Now if I try to merge them I may possibly get a minor increase in the union set (1st term of equation 18). So the overall loss-change will be positive, encouraging to not perform any merging operation.

Can you help me with understanding the optimization scheme?

Thanks in advance!

GaoangW commented 3 years ago

Hello,

Thank you for the question. We have set a threshold for the split operation to guarantee the penalty. Please see TC_tracker/combCost.m at master · GaoangW/TC_tracker (github.com).

At 2021-01-22 07:03:15, "Harsh Nandan" notifications@github.com wrote:

I am trying to understand the tracklet clustering algorithm (section 3.5) of your CVPR paper. I have two questions to start with:

For sake of simplicity, assume that I am just considering smoothness loss for figuring out if two tracklets in their individual set can be merged together.

Is it right to say that if each tracklet is its own set (let's say that is where I start my optimization) the loss of each set will be zero? I am trying to see what is the value of 2nd term in equation 18 of the paper.

Now if I try to merge them I may possibly get a minor increase in the union set (1st term of equation 18). So the overall loss-change will be positive, encouraging to not perform any merging operation.

Can you help me with understanding the optimization scheme?

Thanks in advance!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

harshnandan commented 3 years ago

Thank you, I am going to take a look at the repo you shared.