AMD Aims For AMF Decode In FFmpeg, Questioned Over Vulkan Video Commitment

[Quackdoc]

Yes, but 2/3 aren't on slowpics. Anyways, I've just made a comparison myself, uploading now. No difference whatsoever.

2 of the images are to be compared with eachother, the third link slowpics, contains a set of images. there are 4 comparisons taken at various times throughout the video, each comparison has 3 images. source, svtav1 qsv

[Artim]

To have an actually fair comparison, I downloaded the CC shortfilm big_buck_bunny_1080p_h264.mov. I encoded it with SVT-AV1 and AMD's AV1 encoder on a Ryzen 7 7040 series chip through VA-API in ffmpeg 6.1.1-4+b1 (from Debian's repos). To have the most fair comparison, I let ffmpeg make all decisions and only defined bitrate and maxrate. For 1080p24 in AV1, -b:v 5M -maxrate 5.5M should be more than a fair test. I only encoded 10 sec of the clip as I didn't bother to wait longer. The SVT-AV1 mkv was 8.4 MB, the VA-API AV1 one was 7.6 MB. I then extracted the same frame from the three versions of the clip (at least as far as I can tell it's the same, I didn't bother specifying it too precisely) and of course saved them as png. Here the results:

The original: https://imgur.com/a/50KBzX6
SVT-AV1: https://imgur.com/a/QF0GSM3
VA-API AV1: https://imgur.com/a/9riDbop

[Quackdoc]

What kind of frame was this? if it's a key frame they can look drastically higher quality and more similar then other frames.

but you are also working with a high bitrate in the first place for av1 5500kbps is fairly good for for av1 and thats before counting how badly it was overshot. vaapi over shot some, but svtav1 went wildly overboard

[Artim]

2 of the images are to be compared with eachother, the third link slowpics, contains a set of images. there are 4 comparisons taken at various times throughout the video, each comparison has 3 images. source, svtav1 qsv

Are you really that thick? 2 out of your 3 pics aren't hosted on slow.pics, which results in the impossibility of viewing them at that time was impossible due to the mentioned error message.

And now that they do load, you've just proven me right. Yes, there's a difference, but it's impossible to notice without a side-by-side comparison.

[Artim]

What kind of frame was this? if it's a key frame they can look drastically higher quality and more similar then other frames.

but you are also working with a high bitrate in the first place for av1 5500kbps is fairly good for for av1 and thats before counting how badly it was overshot. vaapi over shot some, but svtav1 went wildly overboard

I'm sorry, that was me missing the control key for copying the actual value. Using -b:v 5M and -maxrate 5.5M was the first trial. The pictures are actually from a second trial where I used -b:v 3.5M -maxrate 3.8M. And to what kind of frame it was: No idea. I didn't pick a specific frame. I saw that at that time code there was a view worth comparing due to all the grass and reflections in the water and had ffmpeg export a frame at that time code to png, only giving the time code in seconds with -ss 00:00:12.

[Quackdoc]

Are you really that thick? 2 out of your 3 pics aren't hosted on slow.pics, which results in the impossibility of viewing them at that time was impossible due to the mentioned error message.

And now that they do load, you've just proven me right. Yes, there's a difference, but it's impossible to notice without a side-by-side comparison.

it's not 2 out of 3, it's 2 out of 14 pictures. the images on catbox are only to be compared against each other as one has significantly more degredation then the other

[Artim]

it's not 2 out of 3, it's 2 out of 14 pictures. the images on catbox are only to be compared against each other as one has significantly more degredation then the other

You may want to check your links again. There are only 3 links and I can only see 1 image per link. So if you expect 14 images to be there, something must have went wrong. While fixing that, you may want to use some more reliable host too, that doesn't refuse to load 90 % of the day.

[Quackdoc]

You may want to check your links again. There are only 3 links and I can only see 1 image per link. So if you expect 14 images to be there, something must have went wrong. While fixing that, you may want to use some more reliable host too, that doesn't refuse to load 90 % of the day.

slow.pics is working for literally everyone who tried

...
After uploading you can navigate between images by clicking on them, navbar or by using keyboard hotkeys.

left and right arrow to swap images, up and down arrow to swap comparisons

[Artim]

slow.pics is working for literally everyone who tried



left and right arrow to swap images, up and down arrow to swap comparisons

Sorry that I expect websites to work properly without JS. Since even with SVT-AV1, image quality is considerably worse and it's questionable if you presets of choice are even comparable, it's very much questionable whether that's a fair comparison even in the slightest.

[Quackdoc]

Sorry that I expect websites to work properly without JS. Since even with SVT-AV1, image quality is considerably worse and it's questionable if you presets of choice are even comparable, it's very much questionable whether that's a fair comparison even in the slightest.

You can take it however you wish. This was for, more or less internal testing that I was doing. I don't really care if people think it's fair or not. It's simply the best I could possibly do with either one of them.

However, when I'm distributing it to people for whatever reason, I definitely know which one I'm going to choose. It is most certainly not a minimal difference.