Video clip shootout!

[stiltman]

I picked clip #2 even though to me they both looked pretty crapy

Just clip 2 looked less crapy

[booaf]

Both clips look crappy if you play them full screen.
If you play them at their target resolution, they look almost the same.
Almost? you may ask?
I say almost, because the XVID clip color space is flawed, because XVID can't properly render the correct colors at such low bitrate.
The mpeg-1 counterpart, although with more visible macroblocks (when you pause the video), has a sharper image that still keeps the correct color space. The XVID has the normal green "bleeding" faults which haunt all XVID and DIVX compression codecs.
I voted for clip #1, because on my Pocket pc it looks better when played at the encoded resolution.
I used this for playback tests http://www.projectmayo.com/projects/detail.php?projectId=9

EDIT: Not to mention that I see the XVID clip has only 4 key frames, and the mpeg-1 has an I frame every 18 frames.
Did you guys miss that
If we lower the XVID key frames to 18, it will look worse that the mpeg-1 counterpart
So this mpeg-1 is pretty damn good, because if it was encoded with a gop of 300 just like the XVID, then it seems the compression will be even far better, and it would be good to see how it would stack up to the XVID.
For a short gop of 18, against a gop of 300 as in the XVID sample, I'd say clip #1 is not bad at all

Greetings,
Baf

[Rookie64]

I had voted very early on - I was probably the second voter, so I had no outside influence in the choice I made.


I voted for the second clip 'cause I could clearly see the visible artifacts in the first clip...the second was a sharper image.


I didn't view at full screen resolution and I didn't need to download the codec, vitualis linked to, to view the clips.

[Cobra]

EDIT: Not to mention that I see the XVID clip has only 4 key frames, and the mpeg-1 has an I frame every 18 frames.
Did you guys miss that

You can't use that to argue a case for kvcd. It is true that it is a limitation of the MPEG-1/kvcd standard, but that's what makes MPEG-4 better.

I voted for XviD. Although it is less detailed that the kvcd clip, the macroblocking is much less obvious and it doesn't have the massive distortion seen on the kvcd clip (like when part of the guy's face end up disjointed). I voted quite early too.

Cobra

[flaninacupboard]

yep, both were crap, clip 2 was just less crap.

I wonder how much we've all been slagged off on the kvcd site...?

[vitualis]

Before another person says "this MPEG-1 is pretty damn good", please step back and look at the clip again.

This MPEG-1 clip is pretty bad. Unwatchably bad.

The XviD clip is rather horrendous too.

Which again brings me to the point that I think this test wasn't fair towards the XviD clip, but I was willing to accept kwag's clips anyway just to see what people thought. This is an absolutely crappy MPEG-4 encoded clip. My experience is that you can do a hell of a lot better than this example at the same bitrate.

As for the number of I-frames, etc., that is irrelevant. MPEG-1 and MPEG-4 work differently. Having a lot of I-frames means absolutely diddly. If I encode a VCD spec MPEG-1 clip (i.e., 352x240 @ 29.97 fps, 1150 kbit/s video CBR) as an I-frame only clip, it will look horrendous against a "normal" one. I-frames are only useful if you have the framerate to support it as it drains bits from every other frame. MPEG-1 and MPEG-4 work differently enough that it is not meaningful to compare the number of I-frames between codecs.

Regards.

[booaf]

Before another person says "this MPEG-1 is pretty damn good", please step back and look at the clip again.

This MPEG-1 clip is pretty bad. Unwatchably bad.

Sorry to disagree with you vitualis. It depends where you play back those clips.
If you look at them on a Pocket PC, both clips will look almost identical.

Which bring me to this question.

You said on another thread --> "For example, try making video clips for Pocket PCs at 320x240. You can get away with reasonable quality video with DivX at around 200-300 kbit/s depending on the video quality. Impossible with MPEG-1 or MPEG-2. "

and then kwag said --> "Here's your PocketPC version of KVCD MPEG-1 at average bitrate of 293Kbps, clearly (in overall quality) beating the CRAP out of XviD"


Well, I think kwag did exagerate a little when he said "beating the crap out of xvid" but I also think you also exagerated when you said "Impossible with MPEG-1 or MPEG-2", as the clip does show it is possible.

The XviD clip is rather horrendous too.

Yes it is, but only when seen on the PC. At it's native resolution, they both are watcheable.
But please, look again at the color space issue.
The xvid colors are flawed, while the mpeg-1 are not
Just open two virtualdub windows, and load each clip.
Do a frame by frame comparison and you'll clearly see what I mean.

As for the number of I-frames, etc., that is irrelevant. MPEG-1 and MPEG-4 work differently. Having a lot of I-frames means absolutely diddly.

I think you misunderstood what I posted.
What I really meant is that the xvid clip has only 4 key frames in the complete clip. If that mpeg-1 had been encoded with a larger gop, maybe a size of 48 or 64, then if could probably match or even exceed the xvid version. That I haven't tested, but it's possible.
But I agree that xvid does work that way, and uses extremely long gops and that's also the reason they achieve large compressions.
However, if you were to encode an xvid with one key frame every 18 frames, and then compare it to that mpeg-1, I'm pretty sure the xvid would be a total disaster, way inferior to the mpeg-1.
But again, xvid isn't supposed to work with such short key frames anyway.

If I encode a VCD spec MPEG-1 clip (i.e., 352x240 @ 29.97 fps, 1150 kbit/s video CBR) as an I-frame only clip, it will look horrendous against a "normal" one.

Of course. I'm aware of that

I-frames are only useful if you have the framerate to support it as it drains bits from every other frame.

Nope. "I Frame" encoding is excelent for high bitrate masterings, as used in broadcast studios. And it's great for editing, as opposed to IBBP encodings, which is harder to edit. "I Frame" encoding is not relevant to framerates. And because there are no forward/backward predictive B or P frames used on "I" MPEG-1/2 encodings, a very high bitrate must be used in order to maintain a high quality.

Greetings,
Baf

[tgpo]

booaf, why aren't you posting under the name "kvcd is the best" pacheco??

[booaf]

booaf, why aren't you posting under the name "kvcd is the best" pacheco??

Where did I say "kvcd is the best"
Why would I post ""kvcd is the best" pacheco"

I'm pretty sure there are other's that think like pacheco. But my opinion was pretty neutral, don't you think ?

Except I said that the xvid clip colors are flawed, and anyone can see that.

So what's your problem

Greetings,
Baf

[Cobra]

Look at the poll results.

Cobra

EDIT - What I mean to say is that the compression artefacts generated by XviD, although present, are more acceptable to the majority of viewers than the artefacts found using the kvcd standard.

[BJ_M]

booaf, why aren't you posting under the name "kvcd is the best" pacheco??

Where did I say "kvcd is the best"
Why would I post ""kvcd is the best" pacheco"

I'm pretty sure there are other's that think like pacheco. But my opinion was pretty neutral, don't you think ?

Except I said that the xvid clip colors are flawed, and anyone can see that.

So what's your problem

Greetings,
Baf

https://www.videohelp.com/forum/viewtopic.php?p=960107#960107

[vitualis]

You said on another thread --> "For example, try making video clips for Pocket PCs at 320x240. You can get away with reasonable quality video with DivX at around 200-300 kbit/s depending on the video quality. Impossible with MPEG-1 or MPEG-2. "

and then kwag said --> "Here's your PocketPC version of KVCD MPEG-1 at average bitrate of 293Kbps, clearly (in overall quality) beating the CRAP out of XviD"

Well, I think kwag did exagerate a little when he said "beating the crap out of xvid" but I also think you also exagerated when you said "Impossible with MPEG-1 or MPEG-2", as the clip does show it is possible.

Um no. I continue to say that it is impossible because this MPEG-1 clips looks like crap.

Yes, you can make crappy looking MPEG-1 clips at low bitrates, no doubt about it.

I said "reasonable" quality... and that MPEG-1 clips is not what I would consider to be "reasonable" quality. To my eyes, it has disruptive artifacts dancing all over the place and I certainly cannot watch a full length movie of it on my Pocket PC. That XviD clip looks pretty bad too. As I stated before, MY DivX5 clips for Pocket PC tend to look much better than both.

I think you misunderstood what I posted.
What I really meant is that the xvid clip has only 4 key frames in the complete clip. If that mpeg-1 had been encoded with a larger gop, maybe a size of 48 or 64, then if could probably match or even exceed the xvid version. That I haven't tested, but it's possible.
But I agree that xvid does work that way, and uses extremely long gops and that's also the reason they achieve large compressions.
However, if you were to encode an xvid with one key frame every 18 frames, and then compare it to that mpeg-1, I'm pretty sure the xvid would be a total disaster, way inferior to the mpeg-1.
But again, xvid isn't supposed to work with such short key frames anyway.

Your bottom line is the key.

And the corollary of that is MPEG-1 isn't really designed to work with long GOPs.

You seem to be under the impression that simply increasing the GOP length (i.e., less I frames and more P and B frames) will yield better compression/quality.

That is a FALSE assumption.

This has been argued time and time before in the older days but I suppose that you weren't here then.

Let us consider the following:

• I frames will tend to yield better quality where there are large changes between two frames
• However, I frames generally require MORE bits than P or B frames as they don't take any advantage of temporal compression
• P and B frames tend to yield better quality (at same number of bits compared to I frames) where there are small changes between two frames
• P and B frames tend to work poorly where there are large changes between two frames.

Now, say we have a video clip with an average of "X" number of P and/or B frames between two I-frames.

For the SAME number of bits (i.e., same average bitrate), would we get BETTER or WORSE quality by increasing the number of I frames?

Argument for BETTER quality:
You tend to get fewer artifacts on the I frames and if mistakes are made by the encoder for motion vector calculations for P and B frames, these frames worsen the video quality.

Argument for WORSE quality:
More I frames mean that you are shifting the bitrate allocation towards the I frames from the P and B frames. I frames are less "efficient" are encoding the video which means in general, the encoding efficiency is reduced. Furthermore, as P and B frames generally have much less bits per frame, more I frames than the original means that the average bits for each I frame has also been reduced. The reduction in artifacts from less P and B frames is overridden by the degradation of the I frame quality.

For the SAME number of bits (i.e., same average bitrate), would we get BETTER or WORSE quality by decreasing the number of I frames?

Argument for BETTER quality:
Overall, P and B frames are much more efficient than I frames. By reducing the number of I frames, the bits previously locked in I frames is redistributed to P and B frames, boosting the overall efficiency of the encoding.

Argument for WORSE quality:
P and B frames, esp in MPEG-1 tend not to handle large changes between frames well (e.g., scene changes). Forcing the MPEG encoder to look for motion vectors between these changes sucks up bits and causes a bad result that can exist for a few frames after this change.

Basically, for any particular clip for any particular encoding method (e.g., MPEG-1) there is no reason to assume at all that increasing or decreasing the "GOP" (i.e., average distance between I frames) will improve or worsen the video quality. Indeed, a change in either direction can BOTH worsen the quality. The reason that MPEG-1 tends to have those standardised GOP lengths is a pragmatic solution. Generally, both longer and shorter GOPs lead to poorer overall quality.

Perhaps you don't know, but earlier in the days of KVCD, kwag made a template that had super long GOPs and was quite enthusiastic about it. As usual, he bashed the rest of us for being stupid for not agreeing with him with the "wink, wink, I'm so clever" attitude. Not long after that, he made a 180 degree turn and dropped that idea and went back to standard length GOPs as per his templates now. I don't have much respect for kwag but he does exhaustively tests his encoding ideas so I think you should take it as a given that his KVCD template would NOT perform better with long GOPs as per the XviD clip.

Which was the point I mentioned initially. You cannot compare the fact that the XviD clip has few keyframes compared to the MPEG-1 clip. The encoding methods are different. I'm sure that bumping up the keyframes of the XviD clip to the level of the MPEG-1 clip will yield a crappy video. However, decreasing the number of I frames on the MPEG-1 clip to XviD levels may well make an even crappier clip too.

Best regards.

[stiltman]

What happens if you don't have a pocket PC? Can I go ahead and use DVD quality instead?











just trying to lighten the mood a little

[Doramius]

I thought both were pretty crappy, but I also considered original source to be the master error. When bringing them down to quality and smoothness at a fully expanded screen, clip 2 just looked better than one. Though both digital forms are on the low spectrum, you can't say the mpeg-1 is better. It would be like saying, "If i record 2hrs. on a VHS cassette at SLP, it will be identical in quality to recording it at SP." That's completely false.

It may not be the best analogy, but I'm sure it gets the point across.