Google Draco

[dr_eck]

Is anyone considering implementing Google Draco in a renderer? It's code for compressing meshes. Here's a link: https://github.com/google/draco

Steve

[papaboo]

What do you mean by 'in a renderer'? As datamodel pre-/postprocessing? Or actually uploading the compressed data and decompress on the fly? (if such a thing is possible)

[dr_eck]

Thanks for responding; I was beginning to wonder if the board was comatose. My understanding is that that Draco is aimed at compressing 3D models. A simple application of this might be to decrease the time it takes to load a large model from storage. It may or may not be possible to use this compression to save GPU memory and/or transfers. My primary aim was just to start a discussion and see what others think of the technology, if it is even on the radar.

Steve

[papaboo]

The forum isn't dead, just really quiet. Like a lion on the hunt.
Personally I can say that Draco's in my backlog and will stay there for the foreseeable future, since it doesn't really help my shoot those precious precious rays.
I doubt that the compression would work for on GPU memory, but if you're interested there are other compressions out there Google Quirin Meyer for a couple of papers on compressing positions and normals.

[toxie]

making the transfer faster is pretty hard nowadays in the age of SSDs and fast GPU connections.
and decompressing on the fly is not the target of the draco library, so for this one needs something else that allows better/faster random access.

[toxie]

Also note this for more info: https://github.com/google/draco/blob/ma ... oder.h#L27
Plus also note that its lossy compression (at least thats what i only saw).

[papaboo]

Well if it's lossy then it's irrelevant. The edgebreaker compression is lossless though. It simply compresses the topology / connectivity and I seriously doubt anyone would consider lossy topology compression. High-res-CAD-model-in, swizz-cheese-model-out would be a no go.

[toxie]

Maybe you can set it to lossless, too? I only saw lossy so far. But it should not lead to holes as long as you base your quantization on the vertices themselves (so no diffs), but again, i haven't looked at what they do exactly in great detail.