a problem about the grasp generation. about dex-net HOT 6 CLOSED

I found the cause of this is. I think there is an inconsistency between the way the SDF is generated(by SDFGen) and the library that is creating the SDF3D python object(Meshpy). The problem can be solved by changing the Sdf3D file in Meshpy. If you go read the init function of Sdf3D, you'll find that at one point one of the transformation matrices is divided by the resolution. If you remove that division the SDF creation will work as expected.
I'm sorry I didn't answer this earlier but the issue was closed so I didn't notice.

from dex-net.

It looks like there is a [50, 50, 50] offset applied to the SDF. Not sure where that is coming from. Can you share the mesh you are using and any of the commands you ran to generate the grasps?

from dex-net.

The mesh is just the 'vase' in example_old.hdf5:

And i just follow the normal process:Open a database;Open a dataset;Sample grasps;Compute metrics;Display grasps for object.
And here are some attributions of this SDF：

Here i have tried other objects in example_old.hdf5, such as bar_clamp,part1 and so on. It is strange that the centers of these objects are [50,50,50]

from dex-net.

Yes, it works. That is the point.
Thanks again!

from dex-net.

hey,how do you do with your own data to create the training set?

from dex-net.

@pyni

from dex-net.