Performance on the BDD110k dataset

dingwoai / FairMOT-BDD100K

A fork of FairMOT used to do MOT on BDD100K Dataset

MIT License

27 stars 9 forks source link

Performance on the BDD110k dataset #13

Open sfarkya opened 3 years ago

sfarkya commented 3 years ago

Hello,

Thank you for providing the code. I haven't seen many BDD results yet. I was wondering if you have the final performance number (MOTA, MOTP, etc) for the Multi-Object Tracking?

Please let me know.

Best, Saurabh

supperted825 commented 3 years ago

Hi @sfarkya,

I am currently also using the repo to train on the BDD100K dataset. The MOT metrics I was able to get after training for 60 epochs is as follows. This is with loading the COCO-pretrained CTdet model with DLA34 backbone. Do you get similar numbers?

long2double commented 3 years ago

Excuse me, it takes a long time to train 30epcoh on the BDD100K dataset?

supperted825 commented 3 years ago

Hi @long2double , it takes around 8 hours per epoch on a single Tesla V100 32GB with batch size 16.