search

kevinlin311tw / ObjLeft

:handbag: Abandoned object detection for visual surveillance
70 stars 25 forks source link

Many false alarms in other videos #4

Open pkdogcom opened 7 years ago

pkdogcom commented 7 years ago

Hi Kevin,

Thanks for the good work! I've tried your code and tested on the given video, and it works as expected. However, when I tried it on other videos, such as the ones on Pets 2006/2007 or AVSS 2007, where it was reported performing well on your paper, there are quite a few false alarms and in most cases it can't reproduce the same results as in your paper (like the ones with pictures in the paper).

I assume I was supposed to change the parameter.txt to make it fit better with different videos and settings. Can you suggest what settings you were using in your experiments with these datasets? Thanks.

lzw19951010 commented 6 years ago

@pkdogcom I have encountered similar problems. At first I thought this was due to the fact that parameters for GMM in this repo is hard-coded with 0.0001/0.002, which is different from the paper(0.0002/0.002). Then I changed the parameters, still I cannot reproduce the same results as in the paper. I was wondering if you have already solved this problem?