Couldn't find matching branch in code about form2fit

[kevinzakka]

Hey @LeThaiSon,

Unfortunately, we haven't provided the script that takes in raw data and processes it into heightmap, masks, correspondences and suction and place points.

To obtain the correspondences, you'll need to apply the rigid transform describing the relative end-effector rotation from pick to place on the object mask. We do this in heightmap space but you could just as well do it in pixel space.

Hope this helps!

[LeThaiSon]

Thank you so much for your immediate respond! Refer to the orignal paper, the matching module is used to establishes correspondences between object picking and kit placement locations. Am I right that it only do pairing correspondense pick and place locations (output from dense pixelwise suction and placement network) or it can do alignment for placing position base on the picking relative position? For example the predict picking position is in center of the object, but the predict placing position is a little bit shift to the left, then without alignment process it can not place the object fit to the box. I think the alignment is calculated in planner (generate p,q value) is that right, but when reading to the planner code I couldn't find something like that. Please correct me if my understanding is wrong! Thank you so much.