Question: Can we only train on the VAVIS and VOS datasets for the second stage?

[RASH1099]

Hello, thanks for your great work!
I have a quick question about the two-stage training setting:Is it allowed to only use the VAVIS and VOS dynamic datasets for training in the second stage?
Looking forward to your reply!

[nickwzk]

Hi @RASH1099, thank you for your interest in our work!

Theoretically, yes, as our training pipeline does not impose any strict constraints on the choice of datasets. However, we have not tested this setting ourselves.
Feel free to try it out, we'd be interested to hear about your results.

[RASH1099]

Thank you so much for your quick reply! I really appreciate it.

I have two more questions to ask for your advice, since I am new to this research area:

1. In your official implementation, did Stage 2 use all four datasets mentioned in the paper for training?
   2.I am using two 24GB GPUs for Stage 2. I reduced the batch size to make it runnable, but the quantitative metrics are not improving, and the loss decreases very slowly. Is this a normal phenomenon?

I would be very grateful for your kind help and guidance.