When is OpenCL exactly being used during the Dense Point Cloud generation?

[Mak11]

Hi everybody,

As stated in the title..When is OpenCL exactly being used during the Dense Point Cloud generation?
As far as I can see it is used at the "beginning" and then the whole "filtering depth maps" is all done on the CPU. Is that right?
Also, are you guys planning on using OpenCL for more computational tasks in the near futur?

Cheers

[Alexey Pasumansky]

Hello Mak11,

Currently OpenCL devices can be used to speed up the depth maps reconstruction process only.

And in the nearest updates OpenCL calculations applied to the other steps cannot be expected.

[Mak11]

Thanks for the reply Alexey,

Bummer about the nearest updates not adding more OpenCL support..but I can live with that for now I guess..just have to be patient when building my dense point clouds (24hrs for the last one...).
Can we expect more OpenCL support maybe in the next 6 to 12 months?

Cheers

[Alexey Pasumansky]

Hello Mak11,

There might be a speed improvement for the depth maps filtering stage anyway, that is currently slow for the projects with the excessive overlap.

[Mak11]

Hello Alexey,

Are we talking about speed improvements using the current CPU path? This would definitely be appreciated even as a "stop-gap" before more OpenCL support in the futur.
BTW here's my dense Point Cloud generation Log:

(PC specs are i7 6700K slightly over clocked, 32GB DDR4, Radeon Fury X w/ 8% Overclcok, Samsung EVO 850 SSDs etc)

[Mak11]

Sorry for bumping this old topic. But I was wondering if 1 year later some progress has been made to move more tasks to the GPUs?

Regards

[Dave Martin]

Mak,

Have a look at the changelog in particular what came in with version 1.3.0 in February this year.

Dave

[Mak11]

Dave,
Yeah I know about those changes as I've using 1.3.x since the pre-release version.
I'm just hoping/wondering if more of reconstruction/matching/etc stages are finally going to been GPU accelerated in the future (like Reality Capture does and which makes it a lot faster than Photoscan but is CUDA only so Photoscan is my only choice as I don't have NVidia GPUs and don't plan to either.). 

Cheers

[Alexey Pasumansky]

Hello Mak11,

In the version 1.3 image matching and depth maps generation steps can be performed on GPU (for NVIDIA using CUDA and for Interl and AMD cards - using OpenCL).

[Mak11]

Alexey,
I know about that..but are you guys working on more tasks being performed on the GPU? (for 1.4 maybe?)

Cheers

[Alexey Pasumansky]

Hello Mak11,

I think that the pre-release of the version 1.4 will be available in the beginning of autumn, so the additional information would be available upon the pre-release.

[Mak11]

Alexey,

Thanks. Will wait patiently now :-)