AssafSinger94 / dino-tracker

Official Pytorch Implementation for “DINO-Tracker: Taming DINO for Self-Supervised Point Tracking in a Single Video”
MIT License
361 stars 39 forks source link

Unable to reproduce the effect when generating dataset/horsejump/masks in preprocess with vitL #13

Closed tengjn closed 1 month ago

tengjn commented 2 months ago

may I ask which model was used to generate the mask? And related setting

tnarek commented 2 months ago

hi @tengjn, we used the last layer of DINOv2-Large (see in the config )