search

JXZe / DualVD

75 stars 12 forks source link

About the triplet #13

Closed Zhang-GK closed 3 years ago

Zhang-GK commented 3 years ago

The triplet is expressed as embeddings. Is there any way to get the textual form? Thanks in advance.

JXZe commented 3 years ago

Thanks for your attention for our work.

As for the label of relation, you can refer Large-Scale-VRD for details.

As for the label of visual object, you can refer Faster-rcnn for details.

What's more, we utilize the relation embedding to capture the visual relationship, rather than the textual label. Since the relation embedding contains more visual objects information, which can be viewed as a more soft strategy to construct visual relationship.

Zhang-GK commented 3 years ago

I get it！Thanks again！