SSD detector doesn't work correctly

[nvnnghia]

Hi, Thanks very much for sharing the code, I run your project using SSD detector, the result I got is very bad, there are many wrong detection. Do you have any ideas? Thank you.

[Nuzhny007]

Hi!
It's the first implementation without testing and other. The SSD detector have some params: https://github.com/Smorodov/Multitarget-tracker/blob/master/Detector/DNNDetector.cpp#L46
I will make API for they but jast now you can set confidenceThreshold 0.3 or bigger.

[nvnnghia]

Thanks for your reply. The problem is not come from threshold. The detected bounding boxes tend to drift to 2 sides of the frames. I think the detector is fine, but there is something happened to the bounding box position.

[Nuzhny007]

Thank you!
It was my error. I fixed it: #66

[nvnnghia]

The problem is still there, the coordinates are changed in a different way. Please check again. Thank you.

[Nuzhny007]

Ok, I need additional info.
I'm starting tracker with params: ./MultitargetTracker ../data/atrium.avi --gpu=0 -e=5 -o=at_ssd.mp4
And result: https://www.dropbox.com/s/ht96sakpafbdeth/at_ssd.mp4?dl=0
I think than it is correct. And you? Can you show video with wrong coordinates?

[Nuzhny007]

And you can see new PR with another example: #67

[nvnnghia]

Thank you very much for your kindly support. SSD detector is awesome now.