Question about Mask

[firekeepers]

Thank you for sharing the job! I wonder how the codes achieve the Mask extraction and use mask to guide cross attention maps I check the demo code but don't find the related code