Open LaurentDelaon opened 1 year ago
Hi Laurent Sure, it's possible to make the learning process configurable (to tune the algorithm behavior). I'm afraid I have no enough spare time to do this right now. Feel free to contribute. Or consider another open source project, for example: https://github.com/zk00006/OpenTLD https://github.com/trippsc2/OpenTracker
Hi @iolkhovsky Can I have your email , I'm willing to fix errors in this repo, but I need a bit of your help in certain things to understand it.
Hi @iolkhovsky Can I have your email , I'm willing to fix errors in this repo, but I need a bit of your help in certain things to understand it.
Hi, feel free to reach me via i.v.olkhovsky@gmail.com
When your TLD program is used on long-time the target detection fail.
a) init a target recognition. b) mask target from camera on long-time. c) re-put the masked target front of camera. the TLD algo is unable to found it, because seem to me all is cleared by learning. Only new image may update base recognition.
This is a miss programming: TLD have long time stabiltiy.
So maybe I suggest that if there is no detection object, it is not necessary to update learning. and also we didn't lose the initial learning and maybe we can redetect correctly apparition. Did you think it's possible to modifiy this ?
Regards Laurent.