1. Join Now

    AVForums.com uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Bug in some professional standards convertors ???

Discussion in 'Projectors, Screens & Video Processors' started by JohnAd, Nov 15, 2001.

  1. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    A linux DVD developer has been having problems with the R2 Simpsons disc.

    The problem he has is very similar to issues I've seen with animation and US shows broadcast here.

    See the section "dealing with framerate conversions"

    http://www.dumbterm.net/graphics/dvd/

    if you look at this picture

    http://www.dumbterm.net/graphics/dvd/simpsons/simp402-top.png

    which is a single field you will notice the double image around Bart and Maggie's mouths.

    Note that this is the data as stored on the DVD so it has not been computer processed (other than decoded)

    My amateur diagnosis of the situation is that this DVD was produced by running the US 60Hz master through a standards convertor which wasn't expecting or doesn't handle film source or animation properly and so go confused and put out a double image.

    If my diagnosis is correct who do we need to tell?

    If not what causes this problem?

    Note that I have seen much worse examples than this, basically any cartoon has this sort of artefact if you use pause randomly quite often you catch a double image. I've also seen this with live action stuff but mostly it has been on programs that looks to me like they were originally done on film.

    Any thoughts from the experts?

    John

    [Opps - edited subject line]
     
  2. RichardA

    RichardA
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    John,

    Nothing like a contentious subject line is there!

    However to put you right - most animation is not at 24 frames per second in the first place - some is as low as 12 some around 16 so this can cause problems in the original telecine transfer - it has nothing to do with standards conversion!

    Other issues that can cause this kind of problem is inadequate pre-processing prior to MPEG compression - errors around small moving areas are a classic MPEG artefact.

    I don't know the exact path of the Simpsons but this stuff is usually on film and would be telecined for PAL directly so there would be no conversion.

    Errors in motion on standards conversion does occur - though only simple systems will give a dual image - the Alchemist is a PHc machine, i.e. Phase Correlated Motion Compensation - this kind of system will create in-between frames from a film source and is fully 2:2 and 3:2 aware so there will certainly not be double images.

    The other thing to check would be that when you freeze the image you are actually freezing two fields of the same frame - the field dominance could easily be wrongly detected.

    There is a big difference between doing a thing, and doing a thing right - not all products are created equal, and there is a heck of a lot more to standards conversion than most of the boxes are capable of doing - The Alchemist is undeniably the best standards converter in the world. period. It's unlikely that this £150000 box was used for the Simpsons and so I'm more than a little surprised at the negative and hostile nature of the title of this thread.
     
  3. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    Richard

    I'm sorry for mentioning S&W in the original thread title. As you said in your PM very unprofessional. It was not my intention to cause offense. The Alchemist is the only one I know, that's the only reason I used it as an example.

    I'm sure the Alchemist is a great product, is it possible to tell us who uses them? Last year I noticed a massive improvement in the quality of NTSC->PAL transfers on the BBC, have you sold one to them recently?

    Do broadcasters use these boxes themselves or do thay farm out the conversion work to post houses?

    If I may return to the problem in hand. Again I don't really know what I'm talking about here but am basing this mostly on what I have seen while doing DScaler testing.

    My understanding is that the Simpsons was shot on film and telecined to 59.94, it is then edited on tape which means that the . 3:2 gets all screwed up and that the final master is on tape. My understanding is that this is quite common practice.

    The Simpsons episodes broadcast here do not seem ever to have 2:2 (or 4:4) pulldown and so my assumption was that they have been standards converted rather than re-telecined.

    As you say some standards convertors are better than others. Is there any way we can pressurize the broadcasters / DVD makers to do things properly and upgrade thier machines?

    You say you doubt the Alchemist was used for the Simpsons. I'm sure you're right but it's disappointing that it isn't? Surely we deserve to get the best transfer possible? If this can be done by taking the original master and feeding through an Alchemist then that is what should be done for all transfers.

    John
     
  4. Stuart Wright

    Stuart Wright
    AVForums Founder Staff Member

    Joined:
    Jan 24, 2000
    Messages:
    13,820
    Products Owned:
    6
    Products Wanted:
    2
    Trophy Points:
    167
    Location:
    Birmingham, UK
    Ratings:
    +9,554
    At John's request, I have changed the title of the thread.
     
  5. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    I see now that the example given above is poor, I'll try and grab a better example from kid's TV on Saturday morning, the cartoons are normally full of this kind of problem.

    John
     
  6. Stuart Wright

    Stuart Wright
    AVForums Founder Staff Member

    Joined:
    Jan 24, 2000
    Messages:
    13,820
    Products Owned:
    6
    Products Wanted:
    2
    Trophy Points:
    167
    Location:
    Birmingham, UK
    Ratings:
    +9,554
    The example you gave very much reminds me of the effect I'm seeing all the time in the mpeg of my daughter's birthday party. Recorded on DV tape with a Sony TRV30 and transferred to mpg on a CD (for distribution to the parents) at around 4,500 Kbits/second. I doubt they are related, but the effect is identical. Frames where edges appear in two places.
     

    Attached Files:

  7. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    Stuart

    Wow, that is similar. So is that a 25fps interlaced DV converted to 25fps interlaced MPEG or did you do anything else to it along the way like deinterlace it convert it to NTSC?

    John
     
  8. RichardA

    RichardA
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    John,

    The example you mention of film converted to 3:2 59.94 Hz and then edited in video could easily cause this kind of problem - the 3:2 sequence will get completely screwed up. I don't know any facilities houses that would actually worry about maintaining a 3:2 sequence for this kind of material.
    If you then take that non-continuous and corupt 3:2 material into an M-PEG encoder (even within a broadcast domain it is likely to be M-PEG compressed somewhere) the long GOP (Group of Pictures) structure can result in big errors in picture content at the mid point between the I frames. This would be enhanced by linear standards conversion, which is the technique used in simpler converters - Generally motion compensated converters are only used on 'worthy' material (i.e. stuff with movement) as the benefits on animation are much smaller.

    The BBC certainly use Alchemist standards converters (their R&D dept. was involved in the early generation of this technology) but the way that the BBC and ITV work means that this improvement is down to facilities houses buying this kind of gear.
    As an example of what can go wrong anyone watching the last two F1 Grand Prix on ITV would have seen horendous motion artefacts - that was a facilities house setting a motion compensated converter incorrectly (yes, it was one of our boxes - but all the motion compensation had been turned off by someone!)

    It would be great if everyone used high quality converters as the picture quality could be improved significantly but, as always, it costs money - we probably have all seen the 'digital' converters to allow NTSC to play on a PAL only TV, or even the ones built into DVD players - they do an adequate job for where they are intended but out top end machines range from £20K to £150K - that's some serious cash to make Maggie Simpson mouth a little clearer, so there aren't enough around. The only way to improve this is to lobby the broadcasters and DVD labels for higher quality conversion - and they'll have to come to us to supply the goodies ;)

    Spectre - there are all sorts of issues converting DV to M-PEG, to the point S&W and Panasonic have worked together to come up with some technology to improve this conversion. The kind of this you are seeing could well be relationships between the compression techniques (and the amount of bit rate you are using!)

    Hope this helps.
     
  9. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    Richard

    Thanks for your explaination. I think I understand the problem better now.

    I think what you're saying is that this is an artefact introduced somewhere along the line by an MPEG encoder putting a dodgy I frame into the sequence.

    Cell Field MPEG
    A Odd P1
    A Even P1
    A Odd I2 <<< Problem
    B Even I2 <<< Problem
    B Odd I3
    C Even I3
    C Odd P4
    C Even P4
    D Odd I5
    D Even I5


    The above is supposed to represent 4 film frames converted to 60Hz and then encoded with video MPEG with 3 frame PII GOPs.

    So the problem comes in because I2 has data from two different film frames but the motion is not easy to detect and this confuses the MPEG encoder and it does the double image thing.

    Once this has happened then there would be little hope of sorting it out however good the standards convertor.

    My only remaining question would be that it surprises me that this problem only effects one field, don't MPEG encoders work on frames or do the professional versions behave differently? (Or have I missed something important?)

    John


    P.S. I can't seem to access www.interpolator.com , it says DNS error (may be a problem at my end)
     
  10. RichardA

    RichardA
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    Good questions John!

    I'll have to ask the MPEG team some of these!

    I've just had a 'quiet' word with our IT people and reported the problem - in the meantime the company website is www.snellwilcox.com (it's the same site as interpolator.com anyway!)
     
  11. vektor

    vektor
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    You're likely recompressing to MPEG-1 using progressive encoding, and your deinterlacer is simply upscaling each field and doing a blur between them.

    So yes, the effect is identical. In the simpsons sequence, I'm convinced the blur fields are a linear interpolations between the two adjacent fields.

    Billy
     
  12. vektor

    vektor
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    I believe your second paragraph is correct. I'll update the website with the Simpsons page to indicate what the frame type is (I,P,B) to help show that these are clearly not MPEG encoding artifacts.

    I'm now pretty convinced about the hypothesis and numbers presented in this email:

    http://www.dumbterm.net/graphics/dvd/on-pal-deinterlacing.txt

    Billy
     
  13. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    Billy

    Hi again, welcome to the AV Forums.

    (Billy is the developer I was refering to in my orginal post)

    I think what Richard is suggesting is that there is the possibilty that the MPEG error occured before the final DVD encode was done, maybe even before the standards conversion was done. The path that each cell takes from artists hand to our screens is a pretty tourtuous one and many problems could arise along that route.

    Also S&W have a great page with lots of interesting documents at

    http://www.snellwilcox.com/internet/reference/books.html

    In particular

    http://www.snellwilcox.com/internet/reference/pdfs/estandard.pdf

    Outlines the role of 3:2 detection in standards conversion. It also fairly clearly describes the state of the art now. It is quite possible that the conversion was done 10 years ago and that we are seeing problems that have already been solved. If that is the case it should be possible to for Fox to redo the conversion and get rid if these problems. However I think there are still a lot of mistakes being made even with good kit as Richard describes above with the Grand Prix.

    Also DScaler has problems with the R1 version of the Simpsons because it appears that the underlying frame rate is 12 rather than 24. This confuses our 3:2 detection code. I might have to put in an animation check into the code this weekend.

    John
     
  14. Mr.D

    Mr.D
    Well-known Member

    Joined:
    Jul 14, 2000
    Messages:
    11,053
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    133
    Ratings:
    +1,130
    digital temporal manipulation is commonly used on series animation to adjust and meet timing changes at the editiing stage: rather than go back and reanimate an entire shot.

    Commonly these can involve frame duping: simple repetition or culling of frames ( can look strobey of fast action)

    frame-averaging (+/- 50% dissolve between two( or more) adjacent frames to stretch or compress frame range)

    optical flow analysis based warping: track recognisable clusters of pixels throughout the frame range : generate vector fields : vary the position of these clusters along the vectors according to where they would be with a revised frame range and warp the original material accordingly. ( normally offers additional motion-blur estimation if speeding up also)

    I'd suspect on animation like the simpsons they wouldn't get anymore sophisticated than a frame average ( which you can easily approximate manually at an online anyway.

    This may well be the root of the artifact.
     
  15. JohnAd

    JohnAd
    Active Member

    Joined:
    Mar 14, 2001
    Messages:
    362
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    18
    Location:
    London
    Ratings:
    +4
    Keith

    If what you say is true then the artefact should appear on the R1 version as well?

    Billy

    Can you let me know at which episode and at what time your samples were taken?

    I'll look at the R1 and see if it has this problem.

    John
     
  16. Mr.D

    Mr.D
    Well-known Member

    Joined:
    Jul 14, 2000
    Messages:
    11,053
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    133
    Ratings:
    +1,130
    Totally worth checking out
     

Share This Page

Loading...