the mask in image and ref_img

[XiaokunFeng]

Thank you for your response. I also noticed this infer function. In this function, the input image and refimg are dictionaries that contain two keys: "image" and "mask". What confuses me is the data format of the mask. Could you please explain this based on the following example?

Suppose I want to perform a video object segmentation task, and the length and width of the test video frames are 224, while the length and width of the reference image are 112 (the entire image is used as a prompt). In this case, how should the "mask" for both image and refimg be set? Thank you.

Originally posted by @XiaokunFeng in #143 (comment)