Do finding the prediction bounding box and finding the prediction segmentation mask coordinates affect each other?

[udkii]

hello. I have three questions regarding the yolact model.

1. Mask R-CNN performs instance segmentation by putting a mask on each bounding box of the object detected by faster R CNN. So, in the case of Mask R CNN, if the bounding box cannot be predicted properly, the performance of segmentation is also degraded.

Does YOLACT first find a bounding box like Mask R-CNN, and then put a mask on it?

As I understand the paper, 'finding the bounding box', 'predicting the class', and 'finding the mask constant' are one set, and 'finding the mask' is like another set. If so, is it possible to build a mask properly even if I am not good at predicting the bounding box?

2. Then, what is the role of the bounding box?

3. And is there a way to extract the segmentation result mask as a coordinate other than an image?

I'd be very grateful if anyone could reply. Thank you for your efforts.