How can i map the predicted rotated mask to the input image?

[AOTY-szy]

I have implemented rotated instance segmentation in my training code, while in inference, how can i change the _do_paste_mask function in mmdet/models/roi_heads/mask_heads/fcn_mask_head.py to map the predicted mask to the input image?

I know converting the rotated boxes to horizontal bounding boxes before extracting mask features is an available solution, but I don't want to do like this.

[zytx121]

We didn't support masks, so I couldn't give some valuable suggestions you. Sorry.

[zhaowei0315]

Hi Experts, I can see 6 values stand for a bounding box, but what the meaning of these 6 values ? thanks a lot.

[yangxue0827]

Can you give me a example?