search

RanaMostafaAbdElMohsen / LMOT

[IEEE Access - 2022] LMOT : Efficient Light-Weight Detection and Tracking in Crowds
MIT License
39 stars 3 forks source link

Comparison against FairMOT Yolo backbone? #3

Open shinaushin opened 2 years ago

shinaushin commented 2 years ago

Do you have timing comparison with FairMOT with Yolo v5s backbone? If they are similar, what makes LMOT more preferable?

Thank you

RanaMostafaAbdElMohsen commented 2 years ago

Hello @shinaushin

Thank you for your question. I did not report the timing comparison with FairMOT with Yolo v5s backbone in my paper. From my experiments, FairMOT with Yolo v5s backbone is faster than original FairMOT. However, the model still suffers from scalability and memory issues, resulting from using Re-ID network. As the number of people in the frame increases, the time taken to associate each detected object with the number of tracklets increases. Therefore, LMOT is more preferable for realtime performance