Some of the input images is just annotated with bbox and some are annoted with bbox and mask So while training it throws error

[MadhuriPatil1694]

for example, there are two instances in an image, a person with polygon annotation, and a bicycle with bounding box annotation, because i care person's shape but don't care bicyle's, only care bicycle's location.
is it possible to train with images like this? if yes how to do that?

Originally posted by @mashuai191 in #1190 (comment)

[github-actions]

You've chosen to report an unexpected problem or bug. Unless you already know the root cause of it, please include details about it by filling the issue template.
The following information is missing: "Instructions To Reproduce the Issue and Full Logs"; "Your Environment";

[MadhuriPatil1694]

@mashuai191

for example, there are two instances in an image, a person with polygon annotation, and a bicycle with bounding box annotation, because i care person's shape but don't care bicyle's, only care bicycle's location. is it possible to train with images like this? if yes how to do that?

Originally posted by @mashuai191 in #1190 (comment)
@mashuai191 were you able to solve this issue

[zhanghang1989]

By default Detectron2 doesn't support this use case. You may change the datamapper to support loading the data and also make changes to the mask_rcnn_loss

detectron2/detectron2/modeling/roi_heads/mask_head.py

Line 32 in cbbc1ce

def mask_rcnn_loss(pred_mask_logits: torch.Tensor, instances: List[Instances], vis_period: int = 0):

[mashuai191]

@MadhuriPatil1694 i tried to change but failed, so i annotated all type's objects finally.

[MadhuriPatil1694]

@MadhuriPatil1694 i tried to change but failed, so i annotated all type's objects finally.

@mashuai191 I converted the bbox coordinates to polygon coordinates, for the images with just bbox annotation and it worked.