Some Questions about model prompt design

[EricLina]

Nice work! However, I have some questions regarding the paper:

1. Since the backbone is shared among multiple task heads, does the training order of different tasks affect the model's performance?
2. The Prompt mentioned in the article is a summary of information from support frames, and it doesn't seem to directly assist in handling different tasks (your prompt doesn't perceive task-specific information). So, how can it be claimed that your model unifies multiple tasks?

Thank you for your reply.