Vegas does not truly smart render MPEG2

PeterDuke wrote on 1/7/2015, 8:20 PM

In this thread http://www.sonycreativesoftware.com/forums/ShowMessage.asp?MessageID=914949&Replies=24 JonnyRoy said, “ I know of no tool that will trim M2T files without re-rendering so it must be hard to do or someone would have done it.”

I always thought that Vegas smart rendered m2t files (HDV), meaning that the GOP(s) where a cut or cuts had been made would be re-rendered but the rest would be passed through unchanged. Certainly VideoRedo does just that. Did JR mean something else?

Also, Laurence said, “Theoretically, a smart-render of HDV and XDcam done at the 1440x1080 dimensions and a 25Mbps bitrate should look the same. That is to say, the bulk of the smart-render should be identical and there should be some second generation loss around the edit points and in to the next I-frame. (etc.) In actual practice, in an HDV smart-render, the degradation around the edit points and to the next i-frame is clearly visable, even to a strongly spectacled individual such as myself. “

I have only done limited testing, but I could not see any obvious degradation at the trims. Either my eyesight is worse than Laurences’s, or I don’t know what to look for (or more likely, both the above!). The nature of the scene may also be important.

I then took a HDV clip and trimmed a few frames off each end and smart rendered it out to a new clip. I then loaded the source and rendered clips into the Vegas timeline, with the source at the top and the rendered clip underneath, and aligned the latter to account for the trimmed frames. I then used the compositing tool to give the difference between the two clips.

Generally the difference view was very dark, but by straining my eyes I could see some noise in the re-rendered GOPs. What surprised me was that I could also see noise from time to time during the “unrendered” section!!!

I then rendered the original clip to a SD MPEG2 file and repeated the exercise with it. Once again I could see noise in the re-rendered GOPs but the noise during the “unrendered” portion was quite blocky.

I trimmed the HDV and SD clips with VideoRedo and in both cases I could see difference noise in the GOP sections, and in both cases no noise during the unrendered sections. (For some reason the HDV clip trimmed with VideoRedo was two frames shorter at the start, but that is another matter.)

So VideoRedo does smart render and Vegas does not, despite the “No recompression required” message and the rapid processing.

Back to post

Comments

PeterDuke wrote on 1/15/2015, 6:48 AM

"Often we're asked to supply camera original footage and it can save a lot of space to simply supply the good takes or trim off the obviously useless bits."

If you want lossless trimming of MPEG2, use VideoRedo or Womble.

Laurence wrote on 1/15/2015, 9:58 AM

I use Womble personally. Another reason being that it also smart-renders any compressed audio. My main reason for smart rendering is to trim down my raw footage. I just do this on longer projects. On short projects I don't bother.

I actually use Womble regularly. The one church I work with has a three camera setup with a switcher and motorized cameras going to a simple DVD recorder. Every week I edit the sermon out of the full service, normalize the audio and trim and fade in the beginning and end, I do this all with Womble. Then I run the Womble smart-render through Handbrake with a decomb filter and mono audio before uploading it to a Vimeo plus account. If I did it in Vegas, I would either have to do a much larger intermediary or frame-serve into Handbrake. Frame-serving would be perfect, but I started doing this long before that was a possibility.

I also edit special performances this way. With Womble, I can take the raw DVD with lots of dead air, lousy automated menus, low audio, and evenly spaced markers every few minutes, and turn it into something way more professional. In a very short time with a combination of Womble and DVD Architect, I can turn out a DVD that has proper menus, normalized audio, and sensible markers. I could actually do the whole thing including the final DVD layout in Womble, but I don't because doing this seems to give me interlace flicker on the static items in the menu graphics. This is only on old CRT monitors, but among older church folk, there are still a lot of these.

Womble is fast and easy and the end result contains mostly video that has never been re-rerendered.

musicvid10 wrote on 1/15/2015, 10:19 AM

Peter Duke,
Very nice test procedure and summary, but for that depth of inquiry, you should be using the MSU quality measurement tool, which is hosted at
http://compression.ru/video/quality_measure/video_measurement_tool_en.html

Using a difference mask is a visual indicator, but not directly quantifiable.

PeterDuke wrote on 1/15/2015, 5:53 PM

"for that depth of inquiry, you should be using the MSU quality measurement tool"

I didn't start out to do an in depth study. It just got out of hand! Thanks for the pointer.

"Using a difference mask is a visual indicator, but not directly quantifiable"

The vector scope is sort of quantifiable. If the scatter is confined to about half the diameter of the smallest circle, the difference is not readily apparent.

PeterDuke wrote on 1/15/2015, 6:03 PM

To study HDV (> 1280x720) I would need the professional version, which is about $1300 AU. Maybe not just now. :(

musicvid10 wrote on 1/15/2015, 9:51 PM

I agree 100% about the vectorcope. It is useful on a few levels.

PeterDuke wrote on 7/3/2015, 5:24 AM

My tests so far in this thread have dealt with HDV. I have now done a few tests with standard definition MPEG2, also with the Main Concept encoder. The source material was DV AVI that had been rendered to MPEG2 with "DVD Arch PAL video stream" template. The render times with force recompression and with smart render are as follows:

Vegas 8: 3:31, 1:53
Vegas 9: 0:44, 0.44
Vegas 10: 4:18, 2:55
Vegas 11: 4:08, 2:56
Vegas 12: 4:08, 2:44
Vegas 13: 5:51, 2:44

Vegas 10 to 13 now show significantly longer render times when smart render is turned off. Vegas 9 is still the clear winner, and smart render off doesn't work, just as before.

The compositing difference function was used to compare the smart rendered video with the source. Vegas 8 output showed no difference except in the last 6 frames. Vegas 9 showed differences in the first 36 frames and last 6. Vegas 10-13 showed differences distributed throughout the clip along with no differences.

Thus no version of encoder is perfect but the 8 and 9 versions are the best. Although differences were visible in the vector scope, it was hard to see any noise in the monitor. The magnitudes of the differences with dumb and smart rendering were comparable.

There is benefit in using smart rendering with later versions of Vegas because of up to half render time and the presence of some frames with no difference.

If you want to truly smart render, use Womble, VideoRedo, TMPGEnc Smart Renderer, etc.

NickHope wrote on 10/21/2016, 1:17 AM

I've been testing this by rendering a timeline of 10 clips of 1440x1080-60i HDV lasting 1 min 43 secs to 3 "smart renderable" formats: HDV, Sony MXF and XDCAM EX. I don't have VP9 or 11. I compared the results in VP13.0e by putting the rendered files on tracks above the original footage, setting the upper track's compositing mode to "difference", and viewing the video scopes.

VP8.0c, Win XP laptop

MainConcept HDV - 1:14 minute render as I had to do this on an old XP laptop that supports VP8. Rendered file is identical to the original apart from up to about 1/2 second at the ends.

VP10.0e, Win 10 PC

MainConcept HDV, Sony MXF, XDCAM EX - Each render took 23 secs. Most of the frames are not the same as the original, but some are, and the 3 renders are not the same as each other. The differences over 1 generation are clear on the scopes but not visually.

VP12.0 b770, VP13.0 b453, VP14.0 b161, Win 10 PC

MainConcept HDV, Sony MXF, XDCAM EX - Each render took 28 secs. Most of the frames are not the same as the original, but some are, and the 3 renders are not the same as each other. The differences over 1 generation are clear on the scopes but not visually. Each format is near-identical (just a couple of brief, minor differences) across these versions of Vegas, but substantially different to the VP10 results. This indicates a change in the rendering technology between VP10.0e and VP12.0 b770, but none since.

When rendering XDCAM EX, "no recompression" was displayed as the render passed the boundaries between the files (unlike HDV and MXF), and the quality of the end sections is indeed generally closer to the original than the HDV and Sony MXF renders.

My conclusion: HDV smart rendering got broken somewhere between VP8.0e and VP10.0e, and has never been fixed. VP10, 12, 13 & 14 all say "no-compression" but are in fact altering most sections of the video.