got-10k / toolkit

Official Python toolkit for generic object tracking benchmark GOT-10k and beyond
http://got-10k.aitestunion.com/
MIT License
559 stars 95 forks source link

VOT test #11

Closed yangkang779 closed 5 years ago

yangkang779 commented 5 years ago

你好,我想请问一下用这个接口的VOT18的评测得到的结果,在baseline 的EAO为0.24+,可是实时的上面才0.20或者0.19,看了一下帧率,40多帧每秒,是实时的算法,结果应该差不太多啊,baseline和实时的EAO得分。谢谢。

huanglianghua commented 5 years ago

Hi @yangkang779, the toolkit only implements AO and A (accuracy), R (robustness) metrics for VOT challenges, and we haven't reproduced the EAO (expected average overlap) metric, which is far more complex (see issue 202 in vot-toolkit).

Probably the 0.24+ you get from this toolkit is the A (accuracy) score instead of EAO. Thanks.

huanglianghua commented 5 years ago

Although your tracker runs at over 40 fps on average, on a few sparse frames it could still be slower than realtime, where a few frames could be skipped and would cause significant drops on accuracy and robustness. Can you check your result time files and find out the frames when your tracker runs below realtime? The performance drops were probably caused by those frames.