Tracker: @loaders.gl/potree about loaders.gl HOT 9 CLOSED

@paujar Thanks for your interest. Unfortunately I am not expecting any movement on the potree format at this time.

• The partial potree implementation was done 5 years ago and not much has happened since then.
• The sense then was that since potree is relatively speaking such a simple format, and we have implementations for the much more complicated I3S and 3D Tiles formats, it would not be much work to complete the support for potree.
• While I'd personally love to see potree supported, I have not been able to get back to it, and there has never been enough interest from any contributor or commercial user of deck.gl to actually make this happen.

If it is truly important to your business, there are teams like Action Engine that have been doing paid work on loaders.gl for years and I am sure they could complete a small effort like this for very modest price.

from loaders.gl.

Interesting stuff! Just to let you know, the binary potree format is about to have a minor update in the following weeks. Instead of a predefined set of attributes, it will be possible to specify arbitrary attributes including name, size, type, description for each attribute. The version attribute in cloud.js will be incremented to identify files with the new format.

from loaders.gl.

@m-schuetz Wow, the potree creator. You started all of this... welcome to loaders.gl!

it will be possible to specify arbitrary attributes including name, size, type, description for each attribute.

That sounds like a great addition to the potree format, thanks for the heads up! It should not be hard for us to support that and maybe also make use of some additional attributes.

from loaders.gl.

@m-schuetz If you are curious about the vision for potree support in loaders.gl, you can see the diagram in the blog post

from loaders.gl.

@m-schuetz We talked to the Hobu / Entwine folks and they indicated you have done a lot of work on advanced rendering techniques for point clouds.

On the deck.gl side we have a good handle on the loading of points, but our PointCloudLayer which is the primitive layer use to render all the point cloud tiles, just does basic 3D rendering.

It does not yet use any fancy point cloud rendering techniques, maybe you have some pointers for us that could help us improve our visuals?

from loaders.gl.

@ibgreen Hi, sorry for the late reply. Regarding visualization, I think that one of the most important and relatively cheap techniques is Eye-Dome-Lighting:

• https://tel.archives-ouvertes.fr/tel-00438464/document (chapter 4)

• https://blog.kitware.com/eye-dome-lighting-a-non-photorealistic-shading-technique/

• https://www.esri.com/arcgis-blog/products/arcgis-pro/3d-gis/eye-dome-lighting/

• EDL in Potree: https://github.com/potree/potree/blob/develop/src/materials/shaders/edl.fs

• Potree's EDL is based on this implementation in Cloud Compare: https://github.com/CloudCompare/CloudCompare/blob/069c9451fd2af5ca1fca9fde1bfe4fe0b14e139d/plugins/core/GL/qEDL/shaders/EDL/edl_shade.frag

Another great algorithm to improve visualization is "High-Quality Surface Splatting on Today’s GPUs". Instead of only rendering the closest fragments, this method blends overlapping fragments together. It gets rid of the worst aliasing artifacts on surfaces, albeit not on sillhouettes. It's also rather expensive so I usually deactivate it by default to accomodate users without high-end GPUs.

• https://www.graphics.rwth-aachen.de/media/papers/splatting1.pdf

from loaders.gl.

Btw, the changes to the file format are probably going to be bigger than initially expected. I'd like to take this opportunity to change it so that it outputs a single file for all nodes, rather than one file per node. It's quite a bit of work to get there, though.

from loaders.gl.

For reference, discussion linked to the potree integration feature: #1927

from loaders.gl.

Is this actually coming to 4.2 then if it was partially implemented in 4.1?

from loaders.gl.