ICCV DeeperAction Challenge - Kinetics-TPS Track on Part-level Action Parsing and Action Recognition Forum

Go back to competition Back to thread list Post in this thread

> video label leak

the ground truth action class of test video can be get according to the video folder, how to prevent participators form using the information to improve performance?
we recommend publishing the performance of video action recognition

Posted by: yuzheming @ Sept. 2, 2021, 12:54 p.m.

I agree.

Posted by: fangwudi @ Sept. 3, 2021, 2:10 a.m.

Pardon me. Did you use the video label during test pahse?

Posted by: fangwudi @ Sept. 17, 2021, 8 a.m.

no, we didn't use the test video label.

Posted by: yuzheming @ Sept. 17, 2021, 8:07 a.m.

thanks for your reply. what's your acc of video classification?

Posted by: fangwudi @ Sept. 17, 2021, 8:24 a.m.

I agree with Yuzheming's remarks that the video label cannot be used, and it is completely possible to obtain a 60.0+ accuracy rate in the test stage through a good trainning model.

Posted by: Sheldong @ Sept. 17, 2021, 3:01 p.m.
Post in this thread