Some confusion after reading papers and code.

[callMeAlexS]

Thank you for sharing your interesting work. It let me learn few-shot semantic segmentation from a another perspective. I have some confusion in reading the paper and running the code, could you help me out?
1.Do I need to retrain the model with RePRI every time I perform semantic segmentation on a new image?
2.Can a Support image set be used for only one Query, and will it be useful for other generic Query categories?