Hello, do I must have to draw the bounding boxes manually? Can I generate the scene if I only pass in an image? Also, if I have multiple bounding boxes pre-detected using GroundedSAM2, how do I pass them into the model?

[Xin-jue]

[Mengmouxu]

In summary, at the input level, we have adopted this interaction logic to facilitate interaction with the model. Moreover, this approach aims to achieve manual selection of objects related to the scene for generation, while filtering out unnecessary objects. If you have this need, you can modify the inference code to achieve automated annotation.