[SOLVED] Cone/cylinder unrolling 4 n00bs

[OMG20Centz]

I'm struggling with something that seems trivial for everyone.

I've read pretty much everything i could find on the forum/wiki about unrolling cylinder/conical meshes.
However i'm getting weird results when i'm applying the transformation.

Input : a subsample of a smokestack, textured mesh


Processing :
1 - RANSAC Shape Detection to determine the cone parameters (results very coherent with other methods)
2 - Unroll using the cone parameters previously determined (see weird results below)
3 - Try not to cry for more than a minute, and start again using a cylinder instead (see weird results below)

Weird results :





Suprisingly, each side complements what is lacking in the other.
As you can see, it's almost good. But still far from being good.

Any hint about what i'm missing ?

[daniel]

It's simply because you do this on a mesh (vertices) and CC doesn't handle the mesh part (the tool only considers point clouds). Therefore the triangles at the separation line are all distorted...

You should clone the (vertices) cloud, unroll it, and eventually remesh it if necessary (with Delaunay 2.5D). Another option is too sample a lot of points on the mesh (with 'Edit > Mesh > Sample points') and then unroll this sampled cloud (you may not need to mesh it then to get nice visuals).

[OMG20Centz]

Yah, isn't it awesome !



Any way to render that pointcloud to an image file ? I mean, other than "Render to file".
Ideally, i'd like to give a GSD and have it geotaged.

[daniel]

You can rasterize the cloud (with the Rasterize tool). Then export it as a georeference raster (geotiff). But you'll only be able to load it in a GIS software.

[OMG20Centz]

The rasterize tool seems a perfect fit, but i can't achieve any better than this :



Although it's an interesting result, I need to keep the RGB components instead of the height colorization, what am i missing ?

[daniel]

Seems you are using an old version. Use the latest 2.8 beta version and you'll have the choice to use the 'RGB' layer.

[OMG20Centz]

Such a n00bie move :=(

Works perfectly with low resolution (10 cm, 1 cm,...), but can't compute the required 8523x45114 image (1MM GSD), despite my 32 Gb of RAM ([ccPointCloud::reserve] Not enough memory).
Lacking a computer with 360 GB RAM, I shrinked my pointcloud into 20 pieces and tried again with these smaller ones, and that runs smoothly.

However, it looks like the "interpolate" function for empty cells is not yet working when the active layer is RGB. Am I right ?



Anyway, i found a workaround by initially subsampling the mesh to a higher amount of points, and that works fine.
Thanks for your guidance in the process

[daniel]

For the records, the "interpolate" function for empty cells now works when the active layer is RGB (or a scalar field). This will be available in the official 2.8 release.