search

GeekAlexis / FastMOT

High-performance multiple object tracking based on YOLO, Deep SORT, and KLT 🚀
MIT License
1.12k stars 256 forks source link

Problem with fastmot on yolov3Tiny #246

Open arielkantorovich opened 2 years ago

arielkantorovich commented 2 years ago

Hi everyone, When I run Fastmot on my network that I train v3Tiny I get very big boundary boxes(Attaches video:fastmot_v3Tiny ), but when I run the same network on darknet everything is fine (Attaches video:yolov3Tiny). It's important to note that this problem doesn't happen when I train yolov4Tiny. The only difference between the networks Is image resize v4Tiny: (420,420) and v3Tiny(608,608) but I change the size in yolo.py I hope someone can tell me what I do wrong?

https://user-images.githubusercontent.com/56262208/150344297-c6bd638a-3f20-4d5c-bc4a-76a33b736357.mp4

https://user-images.githubusercontent.com/56262208/150344371-547368dc-ee72-443c-97bd-ffabc20030b7.mp4

arielkantorovich commented 2 years ago

@GeekAlexis Please help me I see this problem in every model of yolov3Tiny also when I get examples from the internet of v3Tiny and I run FastMOT I get this problem

jaysojitrak commented 2 years ago

you can try 'cfg/mot.json' as per your requirements

arielkantorovich commented 2 years ago

you can try 'cfg/mot.json' as per your requirements

Ofcourse I change cfg/mot.json to my requirements but still only in yolov3 or 3tiny i see this problem its look like the BB multiply by some scale factor.