KDE Apps Now Support QOI Images, Plasma 6 Development Continues

[Nth_man]

When you use a webcam for IR head tracking you have to remove the hotcolor filter (UV/IR blocking filter) from the camera's sensor and put an IR filter in its place to convert it into an IR capture camera. The tape of a floppy disk just happens to work as an IR filter so that's what I used.

Thanks!

[Nth_man]

There are some screenshots in the original article: https://pointieststick.com/2023/08/1...t-continues-2/

[uid313]

Like Zstd, XZ, GZip, etc. Most games store their assets inside various archive formats so the fact that QOI compresses smaller can mean smaller download sizes and less disk usage.

Zstd+QOI could mean smaller download sizes, less disk space used, and faster load times.

Ah, I see. Thanks for the clarification!

[geerge]

The gist of that screenshot is that PNG and QOI alternate in regards to which one has the best uncompressed file size while QOI nearly always compresses to the same size or smaller.

You might get even better results if you use a custom qoi compressor that avoids using the indexed encoding. I did some testing years ago and found that the index op can hurt compression, as long as the compressor is strong enough it often does a better job. Of course that also means that a good chunk of the qoi format is unused and that a custom qoi-like format that did away with index ops entirely would be more efficient, that would obviously break compatibility but in the context of a video game asset where the person compressing controls the decompressor, fair game.

[rmfx]

QOI seems what games needed a long time ago. Does anything beat it at decode/encode speed while producing similar compression?

I think any software that relies on bloated png for icons should switch to qoi. That could only lead to better perf.
And ideally use svg but that's an other debate.

[geerge]

You might get even better results if you use a custom qoi compressor that avoids using the indexed encoding. I did some testing years ago and found that the index op can hurt compression, as long as the compressor is strong enough it often does a better job. Of course that also means that a good chunk of the qoi format is unused and that a custom qoi-like format that did away with index ops entirely would be more efficient, that would obviously break compatibility but in the context of a video game asset where the person compressing controls the decompressor, fair game.

I just went ahead and ripped out index ops from the encoder in this branch ( https://github.com/chocolate42/qoi/tree/no_index_op ), it's still fully conformant qoi format and the encoder should even be a bit quicker as it no longer maintains or checks an index. Here's kodim10 (the png has been optimised with optipng), index is a normal build with index ops and noindex is this branch, compressed with xz:

592787 kodim10.png
652383 kodim10_index.qoi
718167 kodim10_noindex.qoi
503304 kodim10_index.qoi.xz
489004 kodim10_noindex.qoi.xz

So lack of index ops inflates the raw qoi file but reduces post-compressed filesize

[dragorth]

Games already use highly compressed state of the art image compression. That was the whole thing with the PS5 hardware decoder for the SSD, it decompresses the proprietary format made by RAD Game Tools, which I believe was recently bought by Epic. They were the team behind the Blink Video Codec many of us will remember from old PS games from the Blink logo at startup.

Games use compression that works well for the GPU, which is what the BASIS compression system was all about.

This is more about CPU efficiency, not GPU, so I don't believe it will work as well in the Game Engine case, unless you are talking a really small 2D light game, and then as long as you don't play it on a phone or something.

[oiaohm]

l don't know what a compressor is, but I think both PNG and QOI are already compressed. Maybe you meant together with like ASTC and ETC2.

Like Zstd, XZ, GZip, etc. Most games store their assets inside various archive formats so the fact that QOI compresses smaller can mean smaller download sizes and less disk usage.

Zstd+QOI could mean smaller download sizes, less disk space used, and faster load times.

skeevy420 you got so close.

https://docs.unity3d.com/Manual/class-TextureImporterOverride.html

Really its annoying that desktop games have one default and mobile games have a different default. The advantage of ASTC and ETC? and DXT? and BC6H is they have native GPU decode hardware yes they are lossy formats for keeping all the details but games don't notice the difference. Remember the native GPU format means decode on demand and free on demand the decompressed is possible. Yes all the gpu support compression can save on vram usage. Yes using these means you are losing the reason you did the lossless in the first place.

https://www.extremetech.com/computin...re-compression Yes Nvidia working on another compression. Now I do wonder if this is going to be S3 all over again or will Nvidia release the tech without strings attached but this is still a lossy quality. Yes still Lossy compression bits of data is being lost every compression pass so you cannot stack compressing.

There are games that use png images and are not optimized for GPU memory the saving here as clear cut. QOI would for sure give less CPU usage in decode vs PNG but less disk space is a maybe if you note in the benchmark not all cases does QOI compress smaller than PNG so might be a slightly bigger download with slightly more disc usage. Yes the over all average in the maker provided benchmarks has QOI slightly bigger than PNG. So depending on what the bottleneck is and compression results on assets QOI might give no load time improvement over PNG. But lower CPU usage in hand held devices can equal lower heat so CPU staying at higher clockspeed.

Also remember stacking compressing. So archive might be compressing PNG to smaller then QOI might not be able to be compressed so is now stored. Zstd/XZ/GZip will not always compress QOI/PNG images so this is another way the disc saving can up and disappear.

https://docs.nvidia.com/deeplearning/dali/user-guide/docs/examples/image_processing/decoder_examples.html

Do notice Nvidia own image decode when you do a PNG no GPU usage.

One of the reasons why PNG is so insanely hard to implement in GPU is that the design of the format is very single threaded. QOI single page design document does not cover mulit thread/multi shader decoding.

[MrCooper]

Does anyone know how to hide the Wayland OpenGL Renderer window? When it's just the one off game it's manageable, but when it's the game and associated tools you can have 3/5 of those windows...and I dislike auto window grouping so that isn't an option.

It's not clear what you mean by "Wayland OpenGL Renderer window", can you elaborate?

[skeevy420]

It's not clear what you mean by "Wayland OpenGL Renderer window", can you elaborate?

These windows:



They don't even have even actual windows. They just appear in the window switcher and it's always with a Wine/Proton game for me. The ones above are from Orbx and MSFS 2020. While that's happening on I'm unable to click inside an X11 smplayer window started with QT_QPA_PLATFORM=xcb.

I was hoping that someone else here had come across that before. I've seen it asked about on Reddit once or twice over the past two months, but that's it.