Encoder fails for `double` or `float64` ply vertex properties

[ataber]

Hello,

I am trying to compress the following PLY file with the CLI encoder on latest master:

ply
format ascii 1.0
element vertex 8
property float64 x
property float64 y
property float64 z
element material 0
element face 12
property list uint8 int32 vertex_indices
end_header
0 0 0 
10 0 0 
10 0 30 
0 0 30 
10 20 0 
10 20 30 
0 20 0 
0 20 30 
3 0 1 2 
3 0 2 3 
3 1 4 5 
3 1 5 2 
3 4 6 7 
3 4 7 5 
3 6 0 3 
3 6 3 7 
3 1 0 6 
3 6 4 1 
3 2 7 3 
3 7 2 5 

but I get the following error: Failed loading the input mesh: Unknown error..

When I change the x, y, and z attributes to be float or float32, the encoding succeeds, but it fails when they are either float64 or double. Am I missing something obvious? The encoder does support doubles right? Thank you!

[ataber]

Indeed, it looks like position values must be float32 for the CLI.

[ondys]

our ply decoding is still rather lacking, but we certainly welcome any external contribution that would extend it. Adding support for doubles would be doable but note that for now it would be better to convert the values into a float32 attribute as our quantization pipeline doesn't support float64 values at this time.

[ataber]

I made some changes to my fork to support doubles (for posterity looking at this issue). Does your comment about float64 mean that Draco basically truncates internally to 32 bits and then quantizes/compresses?

[ondys]

No, if you supply float64 (DT_FLOAT64) attribute to Draco, it will preserve the values as 64 bit floats. What I was trying to say is that you should truncate it to DT_FLOAT32 before compressing it with Draco, because quantization wouldn't work for DT_FLOAT64 and the compression would be very bad.

[ataber]

Gotcha, thanks for the tip!