Closed linQian99 closed 2 months ago
Hi @linQian99 ,
Unfortunately no. As we point out in our paper a scene of SUN RGB-D dataset is defines by a single rgb-d image, unlike our 6 datasets, where a scene is a point cloud reconstructed from multiple rgb-d images. We think that training a single model on these two modalities is not a good idea. However the architecture itself trained only on SUN RGB-D may work fine, if you tune superpoint clustering properly.
Dear author,
Thank you for your great work! May I ask do you have plan of test on SUNRGBD? Thank you
Best Regards, Lin Qian