search

davheld / GOTURN

Source code for paper: Learning to Track at 100 FPS with Deep Regression Networks, Held, et al. ECCV 2016
http://davheld.github.io/GOTURN/GOTURN.html
MIT License
887 stars 316 forks source link

about multiple object tracking #46

Open littletomatodonkey opened 7 years ago

littletomatodonkey commented 7 years ago

Hello, I was wondering whether the code can be changed to track multiple objects per frame? I tried to modify the code to let it accept multiple objects positions input (for each line in the groundtruth.txt I wrote 8, 16 or 24... numbers to input multi-positions), but then I found the network input is just an image cropped by bounding-box(it is just my understanding...), and I can not find out how to change it.... Do you have any ideas about that? @davheld

umasslowellmj commented 7 years ago

did anyone find solution to this?

davheld commented 7 years ago

I don't think there's a great way of doing this at the moment. You can change the network to accept multiple images as input, but then you'd need to retrain it. I think the best solution at the moment is to track each object in the frame independently; you can change the code to feed the images as a "batch" and output independent predictions for each image in the batch. Anything else that you wanted to do would probably be a research project.