Closed destinyzs closed 6 years ago
In the provided pkls, the coordinate of boxing box is zero-indexed with the order [x1, y1, x2, y2]. The duplicated objects and predicates are removed from the dataset.
The data in test.pkl
was got by faster-rcnn?
No. Is there anything wrong?
So,the bounding box && object labels in test.pkl
are just used to predict the predicates as supplementary information? Am i right?
Yes. For predicate detection, the bounding boxes and labels are given.
All right,get it! Thanks!
Hi,i come again...i have some question about the args.proposal = '../data/vrd/proposal.pkl'
.
What‘s this proposal.pkl means?Is these predicated by fasterrcnn ? @GriffinLiang
The file contains the detection results from the ECCV2016 paper (Visual Relationship Detection with Language Priors).
ok,thanks.
Kongming Liang notifications@github.com于2018年4月12日 周四20:14写道:
The file contains the detection results from the ECCV2016 paper (Visual Relationship Detection with Language Priors).
— You are receiving this because you modified the open/close state.
Reply to this email directly, view it on GitHub https://github.com/GriffinLiang/vrd-dsr/issues/2#issuecomment-380784129, or mute the thread https://github.com/notifications/unsubscribe-auth/Ac1C1JXucwEdyV8bARdaajQ_ZOO0_kISks5tn0U8gaJpZM4StN4n .
@GriffinLiang If i want to apply your model on a new dataset, how do I get the "proposal.pkl"?
Faster RCNN or some other object detection models.
2018-04-12 23:10 GMT+08:00 yangshao notifications@github.com:
@GriffinLiang https://github.com/GriffinLiang If i want to apply your model on a new dataset, how do I get the "proposal.pkl"?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/GriffinLiang/vrd-dsr/issues/2#issuecomment-380839968, or mute the thread https://github.com/notifications/unsubscribe-auth/AGupmtmKebPvHODXm4dxCK1Z_81CH3RXks5tn25pgaJpZM4StN4n .
I have questions about the dataset. The data of vrd in the
train.pkl
andtest.pkl
is different from the ground truth which in thesg_train_annotations.json
andsg_test_annotations.json
,and the amount of these objects and predicates are smaller than the ground truth too. Have you modified the data?Is these bounding boxs&&object labels in thetest.pkl
predicated by fasterrcnn or taken directly fromsg_test_annotations.json
?