search

Wanggcong / Spatial-Temporal-Re-identification

[AAAI 2019] Spatial Temporal Re-identification
MIT License
384 stars 77 forks source link

why is frame avged in gen_st_model.py? #7

Open icefengzhe opened 5 years ago

icefengzhe commented 5 years ago

the avged frame is the time of somebody in center point? but when the person go through one camera twice or more times, the avged frame will be wrong, i am confused. besides, my english is poor, excuse me.

Wanggcong commented 5 years ago

We make a simple assumption that a person appears in a camera only once. But do not be confused by this point, you can use a clustering algorithm to judge how many times a person appears in a camera. I think it can further improve performance in complex scenarios.

icefengzhe commented 5 years ago

got it, thank you :-)

icefengzhe commented 5 years ago

besides, can we use PCB to initialize st_model? then we can use 'joint metric''result to refeed st_model to improve performance. thus, maybe there is no need to mark my data for st_model.

Wanggcong commented 5 years ago

The proposed method includes two streams, you can replace these streams with your method, and use evaluate_st.py for the joint metric.