Morgan MJPEG2000 vs. Cineform

[lordtangent]

Did I get your attention? I've been playing with Morgan MJPEG2000 codec since Ian-T brought it to my attention a couple of weeks ago. JPEG2000 and wavelet based compression has been on my radar for a while now, but I didn't know anyone had done a decent windows codec other than Cineform. I'd like to share my results with everyone since now that I've found my intermediate codec I think other people might be interested in my experiences.

So, first of all, what makes wavelet codecs so great?

1. very high quality for the bandwidth
2. scalable quality vs speed on playback
3. what I like to call "wavelety goodness". When pushed too far, the artifacts wavelet codec DO have are so organic looking the eye doesn't really see them as such.

That's the gist of it. But another thing I like, about MJPEG2000 at least, is that it supports 4:4:4 RGB. (Cineform doesn't except for the "pro" level kit)

I'm not even going to really compare Cineform NEO-HDV and Morgan MJPEG2000 except for in one area: Price. This is because Morgan has pretty much 100% of the same features as Cineform NEO-HDV only it ISN'T crippled in stupid ways like being limited in resolution and color model support.

Cineform NEO-HDV $249 (8bit YUV 4:2:2 only. Limited to 1440x1080)
Morgan MJPEG2000 $30 (8bit YUV 4:2:0, 4:2:2, and RGB 4:4:4 no resolution limit)

Just by the above one liner, if you are like me you probably immediately recognize the potential of Morgan MJPEG2000 vs Cineform NEO-HDV

Since I'm cheap, Cineform was not much of an option at $249. $249 only buys NEO-HDV and it's the lowest end option. (The version of Cineform that actually has the mix of features I want is NEO-2K-Edit and it costs $799!)

From here on I'm just going to talk about Morgan MJPEG2000 and how kick ass I think it is. I've dismissed Cineform NEO-HDV as too expensive at this point.

One of the greatest features of Morgan MJPEG2000, short of the quality, is the fact that it allows for fast "draft mode" type decoding. (Like Cineform) It's basically tune-able to any level required to get real-time performance out of your hardware. Quality goes down as speed goes up. But that's ok for when you are just editing "off line". When you need to do your final render you switch the quality level back to full quality mode, hit "make movie" in your NLE and go to bed. Unless you are making The Lord of the Rings Uber Special Frodo Eats Even More Mushrooms Edition, you movie is probably gonna be done rendering when you wake up in the morning. So, it's fast enough in full quality mode to be practical. (7-10fps) It probably wouldn't be fast enough for editing (in HD) at full quality unless you have a blazing fast machine. But in this online/offline type of workflow it is just fine. I can use full quality in SD res on my machine no problems though, if that gives you an idea of where it stands, speed-wise.

Morgan MJPEG2000 has really crappy documentation so it took me a while to figure out how to use it properly. Basically, it allows you to dial up and down the "quality" on decode by setting how many "layers" it decodes. It works with any VFW software since the setting done system wide for the codec. You use the "Quick Config" panel as your interface for these settings.

The trick is the layers have to be in the media to begin with. You need to make the media with layers, otherwise there wont be any when it comes time to decode. (My mistake in my first experiments was to not make ANY layers. I didn't know what they were and they SOUNDED expensive, to I turned them off) Now I use 10 layers and it seems to be a good number. 10 layers means 10 levels of quality to choose from. On my machine (Athlon 64X2 4600+) I can get real-time 24p from HDV res stuff with 6 of 10 layers. At that level the quality is barely lossless. FAR more than required for just interactive editing. Really, layer 1 of 10 is just fine for off-lining.

I'm attaching a screen shot of the panel. It should make the way the feature is implemented more clear now that I've explained the gist of it. There is a very convenient start bar menu for pulling up this panel. Or you can launch it from the start menu. Setting the quality level at any time is easy. It works brilliantly.



The draft quality video comes in full-resolution to whatever software you're using. It's just a lower quality. But it's lower quality in a nice way, kind of smooth you know... "wavelety"... rather than blocky. Considering that the view port of the "out monitor" is usually low res, it makes no sense to decode the full quality while editing anyway.

The setting named "Levels" provides a different function and that is the number of pyramids / wave domains in the file. Based on my reading about JPEG2000, 5-6 should be a pretty good number of "Levels". I think as long as you don't go with some insane number either way (like 1 or 30) it should be ok to tweak it. I haven't had time to experiment so I don't know how much tweaking this will help or hurt compression. If anyone does learn anything about this, I would love to hear about it.

"Reversible" and "Irreversible" are in reference to RGB to YCbCr color transformation. I use RGB 4:4:4 myself, so this isn't an issue for me personally. But if you are using 4:2:2 or 4:2:0 this setting could effect you. I believe the correct setting to use for normal video (that is, just compressing YCC from a YCC source with no color transform) would be "Irreversible". This is a mode similar to rec709. I haven't tested the YCC modes much to be honest since it's my preference to get into RGB ASAP and stay there until final transcode. (Coming from a CG, compositing background I'm just more confident staying in a RGB world) I can say the YCC modes do work.

My observation so far has been that the compression efficiency of JPEG200 is so great the difference of working in RGB and YCC, in terms of disk space and speed, is totally nominal. This is another reason I'm so in love with this codec.

For the bandwidth, the quality possible with MJPEG2000 is just stunning. I can only just start to see artifacts at 10:1 with the naked eye, and that's flipping back and forth between it and uncompressed. The only way I could really visualize the lossy-ness (even at 10:1) was to setup a comp to compare the two using "difference" and then really boosting the diff with a "levels". Only then was I able to see the magnitude of the loss kind of building up at each successive compression level. It does start to build up after about 10:1 or 20:1 But it doesn't start to get artificial looking until about 50:1 or more. (At least on HV20 footage. Higher quality input footage might not compress so well... more on that in a bit!)

[lordtangent]

I did some compression tests to see what ratio I was really getting compared to pure uncompressed, and to also have some visual reference for just how "lossy" each setting is. (I'm still trying to settle on what a good working compression level is in terms of the trade off of quality vs size) The results are quite illuminating:

## RAW Uncompressed 8bit RGB                                                      
First_Test-0006_24p_RAW_444.avi         1.85 GB (1,995,661,312 bytes)             
                                                                                  
## 8bit RGB Lossless zip type compression (Lagarith)                                       
First_Test-0006_24p_LAGS_422.avi         389 MB (408,641,536 bytes)               
First_Test-0006_24p_LAGS_444.avi         513 MB (538,753,024 bytes)               
                                                                                  
## JPEG2000 compression with various settings (All with 6 "Levels" and 10 "Layers")
First_Test-0006_24p_JP2_LOSSLESS_444.avi 446 MB (468,490,240 bytes)               
First_Test-0006_24p_JP2_5-1_444.avi      429 MB (449,902,592 bytes)               
First_Test-0006_24p_JP2_8-1_444.avi      265 MB (278,138,880 bytes)               
First_Test-0006_24p_JP2_10-1_444.avi     209 MB (219,756,544 bytes)               
First_Test-0006_24p_JP2_20-1_444.avi     109 MB (114,907,136 bytes)               
First_Test-0006_24p_JP2_40-1_444.avi     62.2 MB (65,243,136 bytes) 

Note: As you would expect 4:2:2 JP2 compression tends to be about 25% smaller than 4:4:4 at a give compression level. I didn't have enough data points so I omitted the results

The one most striking things that popped out at me, from this test at least, is how Lossless MJPEG2000 is
1. Smaller than the "zip" type Lagarith Lossless (in 4:4:4 mode)
2. Actually about 4:1 compression over the RAW RGB. (vs the 2:1 you might expect)
3. Only marginally larger than "5:1" (lossy)

It could have just been the source material of course (my guess is the lossy-ness of the original footage makes it more compressible. JPEG2000 compression just "lets the air out" of all that already puffed up data, even in RGB) The same is probably true for Lagarith as well. By all rights it should only be able to get about 2:1 compression of the footage, but it does much better than that. It would be interesting to see the results with live captured via HDMI stuff. The results might be quite different for material that hasn't been though the HDV meat-grinder.

Anyway, the results are all good news! It means even if you want to go full lossless MJPEG2000 you still get reasonable file sizes. AND... you still get all the JPEG2000 goodness. (fast decode mode, etc) JPEG2000 is crazy scalable!

The nice thing about JPEG2000 is that it allows you to set quality as "fixed quality" rather than "fixed bandwidth", so, even though the size of each frame changes a little, each frame gets an equal amount of "love" from the codec. This is precisely the same way Cineform works. So, when I say "10:1" or whatever, really what I really mean is "APPROXIMATELY 10:1" (There is a "quality number" from 1-1000, but I haven't been paying attention to it... it isn't as intuitive relative to what we are used to with other types of compressors.)



And so you can see what these "draft modes" look like, here is layel 1 of 10 layers:

http://hv20.info/yopu/lowest_of_10_at_10-1.png

Here is Layer 6 of 10 layers:
http://hv20.info/yopu/level_6_of_10_at_10-1.png

Keep in mind these are draft modes to get real-time frame rates in HDTV resolution. You can switch up to full quality at any time with only a slight hit in playback speed.

[pascalbrown]

What can I say apart from great effort on your research. This is precisely how the community and users learn and progress. I take my hat off to you.

[Ian-T]

Wow…..this is great stuff....thanks for all of this. I see another sticky. I’m glad you explained the layers because I was confused on how to set them. I want to get the best quality out of my footage. I see you make reference to using it offline (at a lower compression) and then using the best compression in the end. How do I go about using this codec in the HV20’s workflow? So far I have been just running the software through Vegas but without removing pull-down but still loving the results. I want to set it up as an option in the HV20pulldown.exe software just like I have the Lagarith and Huffyuv codecs. I know it’s probably a simple tweak…but…this is me we are talking about :-P

Let me see if I understand what you are saying in regards to the levels. But….when making …say…an intermediate file to work with in my NLE (Vegas) do I compress my footage to a middle ground layer (6 or so) at a level of 6 maybe?…This would/might give me decent real-time editing in my NLE. Once I am done and want to render a final product do I disable compression? If so, do I set the Layers (in the decompression tab) to “0” maybe to get the best possible result?

Also, how do you get your different 10:1…20:1 ratios set? Does that have anything to do with the “Forced Quality” tab? Sorry if I am way off with my questions…but I have been fiddling with this and it seems like I am getting closer in my understanding. Thanks again for your hard work.

[Hallvalla]

Windows only?

[Ian-T]

I believe it is Windows only

[lordtangent]

Windows only?

Morgan MJPEG2000 is Windows only.

Quicktime has JPEG2000 built in, but it doesn't have as many options as Morgans implementation. First of all, it doesn't have the quick decode feature. (Which probably makes it too slow for realtime HD editing on most computers right now) And it's very much a black box. I don't know if it's using YCC or RGB, or if it's doing chroma subsampling. If Apple got serious about supporting it though, it would have more potential than ProRes.

[lordtangent]

@Ian-T

As for the "layers" concept... think of each layer like a packet of quality. The pass that creates them is the last thing that runs in the JPEG2000 scheme. It breaks the frame up into the packets, or you can think of them as chunks, that can be decoded individually later. So if you CREATE your media with 10 layers, that's 10 notches of quality you can choose from when you decode. Decoding every packet provides the full quality of the file, less packets provide less quality of the file.

So, as an example... If you choose Layers: 1 in the decompression dialog, you are telling the codec to only decode the first layer (first notch or packet of quality) The more packets you have the decoder decode, the higher the quality. In the above case, 10 would be the highest.. (since there are only 10 layers in your encoded file) As you mention, setting layers to "0" tells the decoder you want it to decode ALL layers (you don't have to know how many there all... it'll just decode all of them... full quality)

Obviously, the less layers you have the codec decode, the less work the computer has to do. You can tune your play back performance by choosing the number of layers you want to decode. Do you want higher speed to higher quality? The choice it totally yours and you can adjust it based on your need at any given time.

As you have guessed, the check box "Forced Quality" in the Encoder options causes the encoder to use fixed QUALITY per frame rather than a fixed SIZE. This is my preferred mode since getting consistent quality from frame to frame is more important to me than having a fixed data rate. I think most people would prefer that mode. The data rate only fluctuates slightly, so it's a good trade off.

The slider sets the actual "quality", which will provide approximately the compression ratio listed to the right as you move the slider. (or the rather the "quality" from 1-1000 listed in the box just below it.)

"levels" is the number of levels in the wavelet pyramid and should probably be set to something like 5-6 and not much else. (If anyone else has info about this I would like to hear about it.)

[Ian-T]

Thanks..... I was on the right track at least with some of those settings. You just kind of put it in black and white for me. Now that I got a better understanding I'm going to mess with it some more tonight.

Oh...in regards to the HV20Pulldown.exe can I set this codec as an option?

Also, I just want to say thanks to people like yourself, Steve Sudzick(sp), Farnsworth, Eugenia and all other contributers for making this community worthwhile.

[Frank]

Tangent,

Thanks so much for this important research. I got the codec and am trying it out. So far I have found that using lossless reduces my uncompressed avis to 30% of their original size while losing zero quality visually (scaled 1600% view in After Effects).

I am so far not seeing any performance gains when using the various layers in decompression. Since the resolution remains the same even though "quality" goes down, I guess the decode time in my case is less significant than just handling such large file sizes.

Nonetheless, saving 70% on my hard drive with no quality loss makes this codec a terrific upgrade to my process.

Thanks again, and apologies in advance for not responding right away to any reply; I'm headed for the mountains of West Virginia for the weekend.

[tcindie]

I downloaded the demo of version 3 of morgan mpeg2000, but the settings looks quite different, I don't see anything about the layers or anything. Have you worked with it at all, or would I be better off getting ver 2?

[Frank]

I downloaded the demo of version 3 of morgan mpeg2000, but the settings looks quite different, I don't see anything about the layers or anything. Have you worked with it at all, or would I be better off getting ver 2?

Assuming no typo, you need MJPEG2000, not MPEG2000. Also, the demo will not work because it expired already; I had to buy it to try it.

[tcindie]

RIght, I meant MJPEG.. but the settings screen is different than that shown above, it doesn't reference any layers or anything like that. I'm wondering how different it is before I buy it. Even though it's quite cheap, I don't like paying for something that isn't going to work for me.

(See attachment to see the config screen for version 3)

[lordtangent]

@tcindie

Yeah, that screen grab is not the same software I'm talking about. You want "MJPEG2000" it's still at version 2 (2.5 actually) You might have downloaded the wrong demo. Morgan also makes a regular motion jpeg codec. Don't buy that one!

JPEG2000 is wavelet based. It's basically a totally new format and has nothing to do with regular JPEG short of the name. (They should have come up with a new name for what is essentially a new format, but they guys at JPEG don't seem to be too creative in that department) Short of the "Wavelety Goodness" a couple of differences I spot right off the bat from your screen grab are MJPEG2000 has a 4:4:4 mode and a lossless mode. That right there should be enough to get you rev'd up.

@Frank

How fast is you machine? 1440x1080 is a pretty big buffer for anything but a fairly buff machine to work with, even if the decoder is fast. It might also just be an issue of being I/O bound on your HD, especially if you are compressing to lossless (even though the HD doesn't not need to transfer all the data, the codec still needs to seek though the data. If your seek times suck, then you still wont get smooth playback) Have you tried going with lossy? (When you check on "lossless" it is literally lossless encoding, which makes the files larger) I swear, there is no perceptible visual loss even at 10:1, which if you look at my little table should cut down the data rate by at least half again. You should give it a try and see if it's worth the trade off for you.

If you still can't get realtime, an option might be to make "proxy" res files for editing and also high-res files for final render. With Premiere at least, it's easy to switch out media by just changing the path to the media in the file system.

So, for example if you out your media in a dir called:

z:/my_media/LossyVegas_Trip/MEDIA

put identically named proxies in a dir called:

z:/my_media/LossyVegas_Trip/PROXY_MEDIA

Point Premiere to the stuff in the MEDIA dir.

Now by changing the name of the MEDIA you can swtich in and out the proxies

z:/my_media/LossyVegas_Trip/MEDIA becomes... z:/my_media/LossyVegas_Trip/OFFLINE_MEDIA

and z:/my_media/LossyVegas_Trip/PROXY_MEDIA becomes... z:/my_media/LossyVegas_Trip/MEDIA

It's not the most elegant solution, but it works. And you can get smooth playback when editing.

@Ian-T

It's straight forward to make templates for HV20Pulldown.exe but it's hard to describe. SSudzick has some pointers on his site. I intend to share my templates once I'm happy I have them working totally correctly also.

[Farnsworth]

If you still can't get realtime, an option might be to make "proxy" res files for editing and also high-res files for final render. With Premiere at least, it's easy to switch out media by just changing the path to the media in the file system.

Holy crap, that WORKS? I've been searching and searching for a good proxy editing solution for my Premiere/After Effects workflow. So far the best I could come up with was editing with low-res clips, opening the Premiere project in AE (for onlining), importing the high-res clips, and then alt-dragging them one at a time over the low-res versions. Pain in the butt.

But if I can just change the path for the Premiere media, all my problems are solved!

I know this has nothing to do with your thread topic, but I had to express my enthusiasm. I never thought of trying anything that obvious.

[lordtangent]

Farnsworth,

All I know is it worked for me when I tested it. I honestly thought it was a long shot myself, since I assumed Premiere must be storing info about the media like resolution and stuff. But it does work! Give it a try.

The only caveat, which I find annoying, is I couldn't figure out how to change the project resolution. I had to start with the resolution that matched the full-res clip size because I couldn't change after I created the project.

[Indigo]

Very interesting lordtangent. I was thinking of getting CineForm NEO HD for my next project but I'll certainly try Morgan first. You mentioned in another thread being limited to DirectShow prgrams in order to use the variable playback resolutions -- is that sorted out now? Are you using this in Premiere with a choice of playback performance? Another thing I'd like to know: I accept the limitation of 8-bit depth, but NEO HD also changes your HDV from 1440 x 1080 to 1920 x 1080 (square pixels), which is a useful feature. Do you get 1920 x 1080 square pixel output with Morgan?

Great software find...

[lordtangent]

Indigo,

Yeah. Some of my early criticisms of the software were due to me not totally understanding how to use it. (Primarily a result of the zero documentation it has) The variable quality on decode is not a feature that relies on direct show. The sub-resolution decoding DOES. But I've found, at least on my machine (AMD 4600+ 64x2) , I don't need it. Decoding the first few layers when I need top speed is good enough.

That doesn't seam to be the case for everyone though. Slower machines still don't get smooth playback even when decoding just one or two layers. ( Some people here are mentioning they are having problems.) 1440x1080 pixels @ 24fps is a lot to push around in real time, regardless of how light weight the codec is. Even just a slow graphics card could throw a monkey wrench in the works for some people.

Premiere can handle non-square pixels and there is no sense in up-resing your HDV footage until the final render step. (It would just slow things down). I just keep it at 1440x1080

If you want to do your final render to 1920x1080 out to Morgan, it doesn't have resolution limits so it's not a problem. The thing is, you are probably NOT going to be able to play that back in real-time at full quality on anything but the most buff machine right now. Unless you mean to make a high quality archive master, it's not the best choice for your final output. For the last step, I assume most people will be transcoding to something meant for distribution and real time playback (MPEG2 or h264). Morgan MJPEG2000 is meant to be used as more of an intermediate codec, like Cineform.

[Frank]

@Frank
How fast is you machine?

Tangent,

Thanks for your suggestions. It may well be my computer that is the problem.

I have a 2 1/2 year old PC with:
P4 3.2 GHz
4 GB RAM (really only 3 practically)
NVIDIA GeForce 6800 GTO graphics card

I already am using proxies when convenient, but I liked the idea of real-time playback with full quality.

I will try the lossy settings.

[tcindie]

according to your specs your machine should have more power than what lordtangent is running, so it's probably not the computer itself. More likely your windows install is mucked up. That tends to happen after a year or two.

[Worley]

OK, so how do I use the Morgan codec with Vegas? I would like to capture from the HV20, and (maybe) use AVIsynth and VirtualDubMod to rotate the footage, convert to RGB, and then save using the Morgan codec.

Doing such things is completely alien to me, and therefore consider me a noobie and treat me gently.

(I've been trying to use AVIsynth over the weekend, but I was getting errors I didn't understand. Odd, as the first time i used it, it worked... I'll keep at it.)

TIA.

[lordtangent]

Hey Worley,

No problem. Everyone is a noob at one point. And no one can be an expert at everything.

Morgan MJPEG2000 is a regular windows VFW codec, so once you install it it should be available to any software that uses video for windows.

As for all the vDub stuff, I plan to write a quick guide at some point, probably when I release my HV20Pulldown.exe templates for MJPEG2000. Basically, I'll just be describing how I actually developed them. I'm sure a step by step would be handy for anyone who hasn't used vDub much.

I'd like to test my templates a little more before I release them. (they are hardly tested at all at this point) I want to make sure everything works as advertised before I let them in to the wild.

You can see the AVISynth templates thread for the stuff I've released already. (That covers the full-range YCC to RGB conversion stuff)

[Worley]

The conversion to RGB I understand thanks to your excellent work as detailed in a different thread on the forum. Many thanks for that.

I was probably sloppy with the scripts I was trying to write, so I'm not not worried about that at the moment.

What I am eager to know is how to use the MJPEG2000 codec in my workflow. Once I understand what you're doing, I'll feel less of a noobie. I'll just keep reading for now, and it will be a few weeks until there is a pressing need for me to work with HDV and Vegas.

Thanks again.

[lordtangent]

I just added a tutorial on how to make vDub templates in the AVISynth and vDub code thread: http://hv20.com/showpost.php?p=26159&postcount=5


UPDATE!

I've also posted my MJPEG2000 templates for vDub here:
http://hv20.com/showpost.php?p=26164&postcount=6


Enjoy!

[Ian-T]

lordtangent, I have been working with your templates (particulary the 5:1, 8:1 and lossless 4:4:4...and I have noticed what seems to be a problem. I followed your step by step instructions for inplementing the scripts into the HV20Pulldown.exe worklfow. But it seems that on every render (even the lossless) I seem to have been getting a strange noise pattern (especially in dark areas) and a sort of blockiness to the overall video. I have rendered the same file several times and kept getting the same results. When I take the M2t file and the rendered file on the timeline in my NLE there is a big difference in each timeline. I could not figure what was going on. SO then what I did was take the simple script SSudzick put together for Morgan and I manually adjusted the ratios.....the results were perfect everytime.

The MorganJPEG2000 codec does exactly what it is supposed to do. You are correct when you say an 8:1 ratio is a happy medium because to my eyes there was no discernable difference between that and the lossless (though it was about 1/3 the size of the lossless). I don't know if you have seen what I seen with the scripts that you provided but maybe someone else can chime in and give us their feedback. It is as if the video is being compressed way more than what the script is telling us. Thanks again for all of what you have been doing.