aharley/pips2

generalizability of the method

Opened this issue · 1 comments

Hi,

Congratulations on your excellent work and significant contributions. I have a question regarding the zero-shot capabilities of your trained model: do you consider it reliable enough for point tracking in unseen videos?

Hello, I have similar concern about it. I have tried to use the pip2 model on my own video and it could not work well. I directly changed the filename in the demo.py and did not do any tuning.