video adaptation

[zhaowenzhi0033]

If I want to fine-tune a GE-Act model for a new task, according to Figure 7 in your paper, I need to perform video adaptation with video supervision before fully fine-tuning the model with action data. However, I didn't find any example scripts for video adaptation. Could you provide some guidance? Alternatively, is video adaptation not necessary for new tasks?

[ryancll]

Thank you for your attention. You're absolutely right—when sufficient computational resources are available, we do recommend following the approach in the paper by performing video adaptation first. We will include guidance on this in the GE-Act post-training instructions ASAP.

[csccsccsccsc]

Thanks for your concern. We have updated README and clearly outlined the details of the task-specific video adaption stage and the action post-training stage.