search

cvlab-epfl / segmentation-driven-pose

Segmentation-driven 6D Object Pose Estimation. CVPR 2019.
GNU General Public License v3.0
184 stars 37 forks source link

How to make a mesh label for semantic segmentation during training #6

Open 571502680 opened 5 years ago

YinlinHu commented 5 years ago

2D reprojections of the mesh vertex according to the pose can tell you ground truth segmentations.

571502680 commented 5 years ago

But the paper uses the Semantic Separation Grid of S*S，if i use 2D reprojections of the mesh vertex, and its size is the shape of pictures!

YinlinHu commented 5 years ago

Just use the label of the nearest pixel, to generate a rough segmentation.

571502680 commented 5 years ago

Could you share the train.py?

YinlinHu commented 5 years ago

We will release an enhanced version of our method with the training code. (maybe in months)