I was able to reproduce results from 3.2. Open Curve Proposal segmentation and classification, where I kept the T-Net in PointNet as it is. Just a quick question is whether we need to keep T-Net for curve parameter regression. For me it makes more sense to exclude T-Net when it comes to regressing the curve parameters, which should be varying depending on the location of input coordinates. Any ideas?
Hi,
I was able to reproduce results from 3.2. Open Curve Proposal segmentation and classification, where I kept the T-Net in PointNet as it is. Just a quick question is whether we need to keep T-Net for curve parameter regression. For me it makes more sense to exclude T-Net when it comes to regressing the curve parameters, which should be varying depending on the location of input coordinates. Any ideas?