search

nickgkan / butd_detr

Code for the ECCV22 paper "Bottom Up Top Down Detection Transformers for Language Grounding in Images and Point Clouds"
Other
74 stars 11 forks source link

Hi, I have a following question here for butd_cls mode. #46

Closed WeitaiKang closed 7 months ago

WeitaiKang commented 7 months ago 
          Hi, I have a following question here for butd_cls mode.

In Link, you get the detected_class_ids from 'data/cls_results.json', which you mentioned in other issues that it is the predicted class result from pointnet++(pretrained on ScanNet).

However, in the standard mode, where butd_cls = False and butd_gt = False, your detected_class_ids comes from Link. And it is also the predicted class result from pointnet++(pretrained on ScanNet), but with additional DC.nyu40id2class mapping.

I have checked both of the above two detected_class_ids and find that they are difference. Do i miss something here?

ps: I remove the corrupt Link to avoid mistake.

_Originally posted by @WeitaiKang in https://github.com/nickgkan/butd_detr/issues/11#issuecomment-1872337546_

ayushjain1144 commented 7 months ago

Hi,

"In the standard mode, where butd_cls = False and butd_gt = False, your detected_class_ids comes from Link https://github.com/nickgkan/butd_detr/blob/10570e0b6826d4a236b18c2c8fac5903866e1c60/src/joint_det_dataset.py#L572. And it is also the predicted class result from pointnet++(pretrained on ScanNet), but with additional DC.nyu40id2class mapping." these detected IDs are not pointnet++ predictions, but class predictions from the group free detector (we get both detected objects and classes from it).

When butd_gt is true, both of these are replaced by gt objects and classes. When butd_cls=True, then boxes are gt and classes come from pointnet++.

Best Ayush

On Sat, 30 Dec, 2023, 2:37 am WeitaiKang, @.***> wrote:

      Hi, I have a following question here for butd_cls mode.

In Link https://github.com/nickgkan/butd_detr/blob/10570e0b6826d4a236b18c2c8fac5903866e1c60/src/joint_det_dataset.py#L721, you get the detected_class_ids from 'data/cls_results.json', which you mentioned in other issues that it is the predicted class result from pointnet++(pretrained on ScanNet).

However, in the standard mode, where butd_cls = False and butd_gt = False, your detected_class_ids comes from Link https://github.com/nickgkan/butd_detr/blob/10570e0b6826d4a236b18c2c8fac5903866e1c60/src/joint_det_dataset.py#L572. And it is also the predicted class result from pointnet++(pretrained on ScanNet), but with additional DC.nyu40id2class mapping.

I have checked both of the above two detected_class_ids and find that they are difference. Do i miss something here?

ps: I remove the corrupt Link https://github.com/nickgkan/butd_detr/blob/10570e0b6826d4a236b18c2c8fac5903866e1c60/src/joint_det_dataset.py#L616 to avoid mistake.

Originally posted by @WeitaiKang https://github.com/WeitaiKang in #11 (comment) https://github.com/nickgkan/butd_detr/issues/11#issuecomment-1872337546

— Reply to this email directly, view it on GitHub https://github.com/nickgkan/butd_detr/issues/46, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG4OHXQLXNKQ3FCTANKJ73DYL4WKZAVCNFSM6AAAAABBHAB4KKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGA3DANBQGE4DSNA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

WeitaiKang commented 7 months ago

Thank you for your prompt response! I now have a full understanding.