search

longcw / MOTDT

Real-time Multiple People Tracking with Deeply Learned Candidate Selection and Person Re-identification
MIT License
531 stars 130 forks source link

Model Inferencing API #13

Open MathieuTuli opened 5 years ago

MathieuTuli commented 5 years ago

Hi @longcw,

I am trying to use your model in a custom system to do inference on new frames against a gallery of images already seen. I'm having a little bit of trouble deciphering your code and its interdependencies and was hoping you could enlighten me how I could use your model for running such inference.

I was able to build a system that simple takes any video, gets the bounding boxes of people per frame, pass this information to OnlineTracker.update() and display the tracking results. But the new system I wish to build requires different tracker logic and so I need a way of interfacing with the inferencing of your model, so that I can simply get bounding boxes of a new image and compute the similarity of those rois with any other image or image gallery.

Is there any way you can help me accelerate this process? Thanks.

longcw commented 5 years ago

You can find the similarity function in https://github.com/longcw/MOTDT/blob/master/tracker/mot_tracker.py#L265 and https://github.com/longcw/MOTDT/blob/master/tracker/matching.py

MathieuTuli commented 5 years ago

Thanks for the help.