Zhongdao / Towards-Realtime-MOT

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

Only use cuhksysu dataset to train, but when i evaluate on mot16 ,there is no detection #238

Open Xiangyu33 opened 2 years ago

Xiangyu33 commented 2 years ago

hi there, thanks for your work! i just train the model on the cuhksysu dataset, and modify the ccmcpe.json as below

{ "root":"H:/my_datasets/DataZoo", "train": { "cuhksysu":"./data/cuhksysu.train" }, "test_emb": { "caltech":"./data/caltech.10k.val", "cuhksysu":"./data/cuhksysu.val", "prw":"./data/prw.val" }, "test": { "caltech":"./data/caltech.val", "citypersons":"./data/citypersons.val" } }

after 30 epochs , i got a weight. but i run the track.py and test on mot16 train dataset, i got QQ截图20211125212351

could you give me some advice? thanks a lot!

Xiangyu33 commented 2 years ago

by the way,when i train the dataset, the id and nTargets loss didn't drop, i dont know why

Xiangyu33 commented 2 years ago

could you please clarify the meaning of id and nTargets loss? thanks

Xiangyu33 commented 2 years ago

244a12d69d5213b7438d04580125a85 this is training result when epoch is 29