search

xiaobai1217 / RepetitionCounting

Code for "Repetitive Activity Counting by Sight and Sound"
22 stars 7 forks source link

Can't reproduce similar results as described in the paper！ #17

Closed guoxigan closed 1 year ago

guoxigan commented 1 year ago

I tested the model on UCFRep and countix with the checkpoint released by the author, and got lower results. The test results of visual model on UCFRep are 0.604(MAE) and 0.447(OBO)， which far lower than 0.143(MAE) and 0.800(OBO).

xiaobai1217 commented 1 year ago

Hi, you are using the checkpoint trained on Countix for the UCFRep, so surely it cannot reproduce the results. The model certainly should be re-trained on UCFRep.

guoxigan commented 1 year ago

Thanks for your quick reply! I'll try it again.