Vegas Pro 21 Comparison av1 with avc

Comments

Former user wrote on 11/15/2023, 1:50 AM

@Howard-Vigorita Interesting I have one of your lossless clips, I can try that another time.

Also, I messed up with the VMAF and also a differential composite video I posted. As I"m using Vegas less, and I never use the internal codecs I forgot Vegas (internal) doesn't match frame rate of project automatically, so the Vegas AV1 information was completely wrong. I updated the Vmaf and deleted the differential video

RogerS wrote on 11/15/2023, 7:02 AM

I just did new renders of the VEGAS Pro Fanmade Ad with my laptop (see signature) and the render didn't complete- it made it 24 seconds through and then just quit, leaving me with a file that plays fine up until that part. A MagixAVC NVENC render is the full length.

With AV1 the pixelation is still there in my player and VEGAS.

RogerS wrote on 11/15/2023, 8:41 AM

I rendered out just the style transfer section of this project with AV1 and there are numerous glitched frames in my media player:

Next, onto the older sample project (VP 16 sample project). It defaults to "good" project quality. I did a 4K recording and upon close inspection see artifacts:

I then tried a downscale to 1080p. Overall impression is soft. Going frame by frame I saw some artifacts:

Most frames are fine but with scene changes I notice the blocky pattern emerge.

With a piece of X AVC-S footage I often test with I see this blockiness, too:

(zoomed in as most of the frame is fine, it's the low-detail out of focus areas that show it here.

all 5 images are from different frames from a UHD source.

 

Howard-Vigorita wrote on 11/15/2023, 12:47 PM

@RogerS I think you've pointed out a failing of the metrics: they don't seem to be identifying anomalies within frames. Maybe because they analyze, at best, a frame at a time. Could use a glitch detector and counter. Ffmpeg does report more numbers than the frame averages... need to study that to see if one of them represents variance... there seems to be wider differences between the highs and lows for some renders compared to others looking at the plots like @Former user does. Here are the ffmetrics settings I use supplemented by a config setting to disable a new feature that messes with the timing. That setting was provided by the developer in his thread. Next to it is the vmaf plot that shows some rather large variances for Magix av1:

Former user wrote on 11/15/2023, 7:53 PM

I also don't see the artifacts when doing a differential between source Prores422 and Vegas AV1, so either differential doesn't pick it up or in my case It didn't create artifacts in the 1 example I tried.

RogerS wrote on 11/15/2023, 8:48 PM

I just did a render with both sample projects on my desktop, thinking maybe it was a problem with the laptop. Here too the AV1 render did not complete with the VP 20 sample ad project. There were artifacts with both projects, just as on the laptop.

I put the renders and screenshots from it here if you'd like to review or compare to your own renders:
https://www.dropbox.com/scl/fo/n2rv65i65tq6vvxa47bsm/h?rlkey=n85zeg9ka8t0ol1ogbgq04in4&dl=0

I also saw issues just doing a simple transcode (X-AVS S to AV1) so it doesn't seem to be connected to project complexity.

RogerS wrote on 11/15/2023, 10:30 PM

Actually composite mode - difference does seem to be a reliable way to separate source from glitch.

While subtle where you do see edges that are brighter than black do look glitched when I go back to regular visible mode.

If this could be automated it would be very helpful as going frame by frame is tedious.

Howard-Vigorita wrote on 11/16/2023, 12:25 AM

It's also going to reveal what lossy compression means by showing what's been left out in the render. With my reference, seems to be showing mostly faded impressions of whitecaps. I suppose if something big shows up more brightly, that's a glitch. I had to still go though frame by frame because lossless hevc clips may render ok but don't play so good. Might work better with braw v1 which Vegas reads.

GJeffrey wrote on 11/16/2023, 7:42 AM

I have rendered a 3840x2160-50p raw file (link) using internal Magix av1 and svtav1 using frameserver.

1/ I convert the raw file to a lossless magicyuv420 to be able to import it on Vegas timeline

2/ Result

3/ The Vegas av1 file bitrate is very low for such file, hence the blockiness seen. The inability to choose a constant quality or a bitrate doesn't help at all.

4/ The almost bitrate matching svtav1 file gives better objective metrics (also seen during playback) but the result is still not good as such low bitrate. (for example the default svtav1 CRF of 35 rendered a file at double the bitrate)

5/ Rendering through frameserver or directly using the svtav1 encoder gives similar result (higher for the later due to no YUV-RGB-YUV conversion probably).

 

Howard-Vigorita wrote on 11/16/2023, 11:41 AM

@GJeffrey I understand FrameServer might have a new lease on life with v5 free again now that the original developer has it back... I recall seeing a post from @wwaag about that right after I bought the CuminCode version.

Robert Johnston wrote on 11/16/2023, 1:29 PM

Do others see artifacts in sample projects like the to ones in my signature when rendered to AV1? I'll do more testing later this week with both my systems.

@RogerS I definitely see artifacts when I render one of the "titles and text" (Drop Split) to an av1. Sort of like a white highlight on some of the sides of the lettering. Not present when rendered with avc.

Intel Core i7 10700K CPU @ 3.80GHz (to 4.65GHz), NVIDIA GeForce RTX 2060 SUPER 8GBytes. Memory 32 GBytes DDR4. Also Intel UHD Graphics 630. Mainboard: Dell Inc. PCI-Express 3.0 (8.0 GT/s) Comet Lake. Bench CPU Multi Thread: 5500.5 per CPU-Z.

Vegas Pro 21.0 (Build 108) with Mocha Vegas

Windows 11 not pro

RobertoL wrote on 2/19/2024, 9:09 PM

Hello. I am having problems with Magix AV1 encoder. The result of the render shows some unexpected glitches on face and hair of the persons where it loses definition, that does not happen when I render with hevc or AVC. I am using a cpu Intel i7-13700K. The gpu have no encoding capabilities Nvidia gt 1030. I am wondering if someone else have had the same problem or it just me. Thank you.

RogerS wrote on 2/19/2024, 9:15 PM

@RobertoL Don't use the AV1 renderer in VEGAS is my recommendation based on our testing above.

VEGAS doesn't support AV1 GPU decoding at present but that isn't the issue, the problem is in the encodes have blocking glitches every certain number of frames. If you need smaller file size HEVC is a fine option.

mark-y wrote on 2/19/2024, 9:36 PM

I read that it makes videos rendered in av1 small files and with better quality than avc and hevc.

Better quality? Can you provide us with a link to that document or benchmarks? It seems counterintuitive in reference to the past 20+ years of GOP interframe codec evolution.

Smaller files, yes. Compression trickery is the game, and smaller files are the payoff. Quality tradeoff is often the casualty; I don't recall anything that would suggest a quality advantage in the historical progression from MPG->AVC->HEVC->VP9/AV1.

GJeffrey wrote on 2/19/2024, 10:55 PM

I read that it makes videos rendered in av1 small files and with better quality than avc and hevc.

Better quality? Can you provide us with a link to that document or benchmarks? It seems counterintuitive in reference to the past 20+ years of GOP interframe codec evolution.

Smaller files, yes. Compression trickery is the game, and smaller files are the payoff. Quality tradeoff is often the casualty; I don't recall anything that would suggest a quality advantage in the historical progression from MPG->AVC->HEVC->VP9/AV1.

@mark-y

FYR, objective metrics with a 420 8bits UHD source

Newer codec are better at low bitrate. With high bitrate, no big differences

mark-y wrote on 2/19/2024, 11:08 PM

Newer codec are better at low bitrate. With high bitrate, no big differences

Yeah, we know that. What that proves is that about the same quality can be achieved at at a lower bitrate with newer compression schemes.

But better? That is a pretty remarkable claim, one that I haven't seen, and so my question. AVC is not inherently better quality than MPG, HEVC is not inherently better quality than AVC, so if AV1 is significantly better quality than any of those at the same bitrate, I would love to test it, as I've done with all the others at some point...

That said, the NVENC numbers you showed are pretty impressive.

Former user wrote on 2/19/2024, 11:21 PM

@GJeffrey That's one of the software AV1 encoders I believe, that's not something people would typically use, You want results for NVENC, Intel, and AMD AV1 (if AMD has AV1) . I have tried the AV1 NVENC encoder and was a bit annoyed that with a 1080P60 file the quality kept going up the higher the bit rate, I think something like 150/100/50mbits. I had hoped to see the quality plateau at much lower bitrates for 1080P60 AV1 .

That was with Vegas and voukoder some time ago, I could check again some time