search

SilvioGiancola / TrackingNet-devkit

Development kit for TrackingNet
https://tracking-net.org/
170 stars 32 forks source link

About the pretraining experiment for SiamFC in the paper #11

Closed jimmy-dq closed 4 years ago

jimmy-dq commented 5 years ago

HI, Dr. M¨uller, Thanks for your great work. I am confused that how do you select one of the twelve chunks (2511 videos) to retrain SiamFC? Do you randomly select one of the twelve chunks videos in each class? Looking forward to your detailed description.

thias15 commented 4 years ago

Hello @handworker.

We just picked the first junk out of the twelve. The idea was just to show the benefit of the additional data. We would expect even better results if more junks are used.

jimmy-dq commented 4 years ago

Hi @thias15 , When I download the dataset, I find the answer. There are several chunks in the data. Many thanks for your reply.

bamboopu commented 1 year ago

Hello @handworker.

We just picked the first junk out of the twelve. The idea was just to show the benefit of the additional data. We would expect even better results if more junks are used.

Thanks for your sharing. As a personal researcher, I was scared by the scale of the dataset for many years because of my limit gpu. Now I know that only the first junk was used.