search

Zhongdao / Towards-Realtime-MOT

Joint Detection and Embedding for fast multi-object tracking
MIT License
2.37k stars 539 forks source link

The MOT16 train set results in README.md can not be reproduce? #215

Open nemonameless opened 3 years ago

nemonameless commented 3 years ago

We just got MOTA of 73.1 70.1 63.5，can not reproduce your results MOTA of 74.8 70.8 63.7 in README.md. Please give more details.

adi1999 commented 3 years ago

hey @nemonameless , just wanted to ask that whether you're able to reproduce results on MOT16 dataset, becuase I'm facing some minor error( UnpicklingError: invalid load key, '\x00') while using pretrained darknet model. If yes, did you had to convert the dataset into the format of MOT16 - images ( jpgs) - labels ( txt files), or have you just tested on original format of MOT16.

nemonameless commented 3 years ago

@adi1999 Just use the data format of this repo, the MOT results of MOT16-test in this repo can be submitted to the https://motchallenge.net , got MOTA 64.4 (1088x608). But for MOT16-training, we cannot reproduce the 74.8 of README.md in this repo.

Curry-Christopher commented 3 years ago

We just got MOTA of 73.1 70.1 63.5，can not reproduce your results MOTA of 74.8 70.8 63.7 in README.md. Please give more details.

I also got MOTA of 73.1 of JDE-1088x608 model, have you found the reason?