3DSSG evaluation

[choibigo]

Hello, i'm interesting your paper(Learning 3D Semantic Scene Graphs from 3D Indoor Reconstructions)

i have a question about evaluation

i just want to know difference between Multi Predicate, ObjCls from PointNet Features and Multi Predicate, ObjCls from GCN Features

you wrote this sentence that The last part of the GCN consists of two MLPs for the prediction of the node and predicate classes So I wonder why your model is not the last row.

Thank you for reading