about segmentation network about pointnet

[sheromon]

I recently started playing with this model. (BTW, thank you @charlesq34 for this excellent release. It is much higher-quality and easier to use than other projects that I've looked at.) @sitzikbs and @touristCheng, there are programs such as MeshLab that will recognize the .obj file format. Using MeshLab, I just imported the "pred" (prediction) and "gt" (ground truth) .obj files separately, and you can compare them. I imagine that other programs for 3D visualization will do the same.

[daerduoCarey]

The one-hot vector is there for fair comparison to the other baseline methods which target at segmentation for shapes from a given same object category. Our part-seg PointNet can work without one-hot vector but there will be some performance drop, around 1-2%. The fact that our PointNet can train across different object categories leads to both positive effects (e.g. easier to train, no overfitting to small subcategories, can be easily finetuned to new shapes) and negative ones (e.g. some segmentation labels may be from different object categories).

Thank you.

Bests,
Kaichun

[touristCheng]

Thanks very much! I want to know where I can get the original 3D models for sampling the points, and how can I visualize the segmentation results, thanks for your help.

[sitzikbs]

@touristCheng If you are reffering to the classification it is in the ModelNet40 website. If it is to the segmentation then I think it is in the shapenet ( not sure though). The code has a file to download the sampled data.

I too will appreciate a visualization code for the segmentation results ( Im sure there is a file like this somewhere since the paper shows these results)

[sitzikbs]

Thanks @sheromon. I already wrote some code to visualize it. Hopefully Ill upload a branch to this in a few months.

[charlesq34]

seems the issue has been resolved. closing for now.

[pournami123]

@sitzikbs
Have you written the code for visualization?

[sitzikbs]

@rejivipin
try the functions here :
https://github.com/sitzikbs/3DmFV-Net/blob/master/utils/visualization.py

[noridayu1998]

@rejivipin
try the functions here :
https://github.com/sitzikbs/3DmFV-Net/blob/master/utils/visualization.py

@sitzikbs
this is for visualizing the semantic segmentation result?

[sitzikbs]

The function visualize_pc_seg should work but I am sure that there are better implementations by now.

[noridayu1998]

The function visualize_pc_seg should work but I am sure that there are better implementations by now.

@sitzikbs
the function visualize_pc_seg will export image file. there is possible to change it in point cloud file?