Fast image processing library #65720

New Issue

Richard Antalik · 2019-06-11T14:11:53+02:00

Richard Antalik commented

2019-06-11 14:11:53 +02:00

Performance of many image operations used in video sequencer and other modules can be improved.

This can be done by using intrinsics functions, prarallelization or using GPU for processing.

Challenges to overcome are mainly HW and SW platform diversity. By doing quick research, use of Vulkan seems to be best option here.

Performance of many image operations used in video sequencer and other modules can be improved. This can be done by using intrinsics functions, prarallelization or using GPU for processing. Challenges to overcome are mainly HW and SW platform diversity. By doing quick research, use of Vulkan seems to be best option here.

Richard Antalik commented

2019-06-11 14:11:53 +02:00

Added subscriber: @iss

Jeroen Bakker commented

2019-06-11 16:48:12 +02:00

Added subscriber: @Jeroen-Bakker

Jeroen Bakker commented

2019-06-11 16:48:12 +02:00

Hi @iss

What are the requirements of such a library. I would expect that headless rendering is one of those.
Did you have a look at Halide? https://halide-lang.org/
They promote them as a language, but actually is more a framework.

BTW I am totally not aware what the future target of the VSE is, but this seems to be a large undertaking when you want to select something that is also for other areas (like compositor etc)

Hi @iss What are the requirements of such a library. I would expect that headless rendering is one of those. Did you have a look at Halide? https://halide-lang.org/ They promote them as a language, but actually is more a framework. BTW I am totally not aware what the future target of the VSE is, but this seems to be a large undertaking when you want to select something that is also for other areas (like compositor etc)

Richard Antalik commented

2019-06-11 18:17:42 +02:00

@Jeroen-Bakker
Thanks for reminding me of halide - forgot about that one.

What are the requirements of such a library.

From VSE perspective it could be quite simple. Just basic image operations on raw data would be OK. Compositor nodes doesn't look that much different from what I would expect from such library. Question would be more about how to handle Blender specific controls, or splitting image to multiple components for example.
Advantage using existing universal library is, that it is already done, tested and maintained.

Using frameworks/CL we get more freedom to do case-specific optimizations, but we have to develop and maintain processing algorithms ourselves, but still we don't have to think too much about OS/HW.

I don't think it is really apropriate writing low level code such as CPU dispatchers in this case.

I am totally not aware what the future target of the VSE

Me too :)
My goal is to review current features, and make sure, that they work correctly and efficiently.

but this seems to be a large undertaking when you want to select something that is also for other areas (like compositor etc)

I am fully aware. That's why it's "long term project" :)
This task is rather idea placeholder, then design document. I don't expect to start working on this for next 2 releases at least.

@Jeroen-Bakker Thanks for reminding me of halide - forgot about that one. >What are the requirements of such a library. From VSE perspective it could be quite simple. Just basic image operations on raw data would be OK. Compositor nodes doesn't look that much different from what I would expect from such library. Question would be more about how to handle Blender specific controls, or splitting image to multiple components for example. Advantage using existing universal library is, that it is already done, tested and maintained. Using frameworks/CL we get more freedom to do case-specific optimizations, but we have to develop and maintain processing algorithms ourselves, but still we don't have to think too much about OS/HW. I don't think it is really apropriate writing low level code such as CPU dispatchers in this case. > I am totally not aware what the future target of the VSE Me too :) My goal is to review current features, and make sure, that they work correctly and efficiently. > but this seems to be a large undertaking when you want to select something that is also for other areas (like compositor etc) I am fully aware. That's why it's "long term project" :) This task is rather idea placeholder, then design document. I don't expect to start working on this for next 2 releases at least.

Richard Antalik added this to the Video Sequencer project 2023-02-09 20:57:27 +01:00

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Fast image processing library #65720