XHA1 & Premiere quality issues. NEED HELP!

[RPK]

Hi,

I have been struggling getting high quality rendering from Adobe Premiere CS3 and CS4 on both PC and OSX. I've read a gazillion threads on various workflows and have tried out a lot.

Here are sample areas from two different stills at 100% zoom. The stills are each shown as exported from Final Cut Pro in comparison to the lower quality with artifacts when exported from Premiere CS4. Horizontal pixelation is clearly visible around hand area and in the red areas of the microphone.

FCP export:

CS4 export:

FCP export:

CS4 export:

I'm capturing 24f footage either via XHA1 or HV20 through firewire.
Among the methods I tried are direct capture to Premiere with included presets (HDV 24p) or the old Canon 24f Premiere presets. I also tried HDVSplit. Captured with the latter the footage looks fine when viewing in Windows Media Player, but I get the same artifacts once the footage is in Premiere. Also tried capturing in CS3 on OSX with the same problems.

I captured the same footage in FCP and the results look much better! However, I much (MUCH) prefer Premiere on PC over FCP on Mac, but that's enough headache for another thread.

I would be very glad for any help with getting the quality right with Premiere CS4!

Thanks,
RPK

 

[strivedi]

Are you getting these issues only when you export stills? Or do you see the same issues in the source monitor when playing video?

I have seen quality degradation when exporting stills from Premier Pro. It is not as good as the HD video footage and certainly not as good as a still camera.


Looks like the tabla footage is in low light .

What preset are you using on the camera? That could make a difference in the source footage.

 

[PDR]

I also tried HDVSplit. Captured with the latter the footage looks fine when viewing in Windows Media Player, but I get the same artifacts once the footage is in Premiere.

If the artifacts appear when importing into premiere, putting footage on a timeline (i.e. visible in the preview even before exporting or rendering), it's likely a mismatched interpretation of the footage and sequence settings (e.g. pulldown not removed correctly)

If the preview is fine (on high quality settings), and only the export is lower quality, you have to provide more information on your export settings used

 

[KeeganFlynn]

I'm with PDR, are you having problems when exporting? or on capture?

Getting a master quality file out of FCP is a breeze, but to do the same out of Premiere can be a bit of a pain if you're editing in native HDV.

 

[RPK]

Thanks for the replies! I really appreciate your help!

I'm with PDR, are you having problems when exporting? or on capture?

The artifacts appear as soon as the footage is inside Premiere, visible in the Program Monitor and when viewing on 'external hardware' (might not use accurate Adobe lingo here.) This is before exporting but after capture in HDVSplit! And it's visible in both stills and video rendered in various formats and quality settings.

When using the 'external hardware' option with a Cinema Display (1920x1200) while editing the footage looks like in the examples. However, to my surprise, the footage looks even worse when viewed in the Program Monitor, even with settings at 'high' and zoom at 100%. In the Program Monitor I get artifacts as if the zoom factor isn't at 100%. In FCP the preview monitor with high quality setting and at 100% zoom shows footage just like exported video/stills with high quality settings.

Looks like the tabla footage is in low light .

What preset are you using on the camera? That could make a difference in the source footage.

Yes, the tabla sample is from low light footage, but the artifacts are the same on all kinds of footage.

I'm using a tweaked preset, but the footage looks as it should in FCP and before importing into Premiere, so the preset shouldn't have anything to do with it.

If the artifacts appear when importing into premiere, putting footage on a timeline (i.e. visible in the preview even before exporting or rendering), it's likely a mismatched interpretation of the footage and sequence settings (e.g. pulldown not removed correctly)

I think you are probably right about an error in the footage interpretation. I will take a look at the settings in Premiere and will report back here. However, since I'm working with 24p footage from the XHA1 there shouldn't be any pulldown issues, right?

Thanks again for your help!

RPK

 

[RPK]

it's likely a mismatched interpretation of the footage and sequence settings

OK, here is what Premiere says in the 'Interpret Footage' dialog with footage captured directly in Premiere (not HDVSplit):

- Frame Rate: 23.9760
- Pixel Aspect Ratio: 1.333
- Field Order: No Fields (Progressive Scan)
- Nothing checked in Alpha Channel.


The sequence settings are:

-HDV 1080p, 23.9760
- 1440x1080, 1.333
- No Fields
- 24fps Timecode
- Preview file format: I-Frame only MPEG
- Unchecked: Maximum Bit Depth & Maximum Render Quality.


Looks like everything matches. Placing check marks in Maximum Bit Depth and Maximum Render Quality didn't improve the artifacts.

RPK

 

[PDR]

Do you have the lastest updates for CS4? I think 4.1 added proper pulldown removal for some cameras

Is there a red render bar when you drop the footage onto timeline (if there is , this suggests some mismatch)

However, since I'm working with 24p footage from the XHA1 there shouldn't be any pulldown issues, right?

I don't have this particular camera, but does it shoot 24psf or 24pA ? It's normal to see the artifacts, until pulldown removal or the footage is interpreted correctly. The 24p native footage is packaged in a 60i stream (if it was 24pN, it would be native)

Were there any frame drops during capture? => can lead to cadence swap , you would have to re-capture

Your sequence settings appear correct to me

 

[RPK]

Do you have the lastest updates for CS4? I think 4.1 added proper pulldown removal for some cameras

Yes, I'm running 4.1.

Is there a red render bar when you drop the footage onto timeline (if there is , this suggests some mismatch)

No, no rendering necessary. No colored bar at all.

It's normal to see the artifacts, until pulldown removal or the footage is interpreted correctly.

I searched for pulldown removal issues in CS4 and found this thread on the Adobe forums: http://forums.adobe.com/message/1946437

However, my 'Interpret Footage' settings are correct and CS3 never displayed or rendered my footage from the XHA1 any better than CS4. Clips are also correctly identified by CS4 as having a 23.976 frame rate.

I'm wondering if this is really a pulldown removal issue.

Windows Media Player displays the captured clips in perfect qualityrogar-Mark-01(DBG)

RPK

 

[PDR]

Windows Media Player displays the captured clips in perfect qualityrogar-Mark-01(DBG)

This doesn't provide much information, because WMP relies on system installed directshow codecs. So the specific splitter and directshow decoder may or may not be post-processing it in the directshow chain. This will vary between systems.

To me, this is highly suspicious of a pulldown removal issue. Many Canon cameras have issues with the 24p in 60i format (24psf). The combing pattern in your screenshots looks a lot like it. The treatment for 24pA is different.

Footage from other Canon HDV cameras exibit this same behaviour. There are dozens of theads on this at the HV20 forums, including workarounds with avisynth to remove pulldown.

If you can't figure it out, if you upload a sample to a free hosting site (e.g. mediafire.com), I can have a look at the footage , and maybe suggest some options

You can check the HV20 forums too, I think there are a few with your camera there

EDIT: not sure if this will work but here is a preset which may interpret it correctly
http://www.adobe.com/support/downloads/detail.jsp?ftpID=3408

 

[RPK]

Hi PDR,

Thanks for the offer! I just sent you a PM with the location of an uploaded file.

EDIT: not sure if this will work but here is a preset which may interpret it correctly

I tried these presets a while ago and they didn't fix the issues.


RPK

 

[PDR]

This is definitely an interpretation issue. The footage as you provided is 24p HDV with standard 2:3 pulldown, but there are 8 field repeats in the very beginning. This will likely throw off most editors because of the cadence break, unless they have adaptive field matching capabilities. The rest of the footage is standard 2:3. There are flags in the stream that indicate 2:3 should be used, but Premiere for some reason seems to ignore this; you said FCP worked fine, I suspect it acknowledges the flags correctly.

It would be nice if premiere supported this type of footage natively (there would be no red-render bar), but I don't know of any tricks. You might search the HV20 forums for new tips & tricks.

The workaround I would use is avisynth with the premiere avisynth import plugin.
1) Install avisynth

2) Install Premiere AVS Plugin http://videoeditorskit.sourceforge.net/

3) Install ffmpegsource2 (The contents go into the avisynth/plugins directory) http://code.google.com/p/ffmpegsource

4) Write the following in notepad in the same directory as your video, changing filenames to match. Save it and rename the extension from .txt to .avs

FFMpegSource2("sample.mpeg", atrack=-1)

5) Import the .avs file and conform (Interpret) the imported .avs as 1.333 PAR (.avs scripts are interpreted as 1:1 square pixel). The .avs is like a proxy that represents your footage, with pulldown removed

There are other methods in avisynth as well using DGIndex, TFM, TDecimate and using adaptive field match, but this is probably the easiest way and it works on your sample.

There are avisynth batch scripters if you have multiple files

It would be waaay easier to use FCP if it interprets the footage correctly

This is a known issue with that camera and similar canon cameras, a quick search:
http://www.dvxuser.com/V6/showthread.php?t=110397
http://www.dvxuser.com/V6/showthread.php?t=117029

There maybe 3rd party utilities to remove the pulldown easily, like cineform

Cheers

 

[RPK]

Wow, thanks so much for taking the time to look at the footage and the detailed help!

I have a few questions:

It would be nice if premiere supported this type of footage natively (there would be no red-render bar)

This footage does not result in a red render bar when I import it and Adobe states that they do support XHA1 24f footage natively. I guess I'm wondering if other Premiere/XHA1 users simply ignore the artifacts (which are admittedly not visible much when exporting to DVD or web resolution) or if I could use another way of capturing so that I don't get those field repeats. The combination of XHA1 and Premiere should be very common, and I'm surprised there aren't more threads about this issue and potential workflows, just like there are for HV20 footage.

1) Install avisynth

2) Install Premiere AVS Plugin http://videoeditorskit.sourceforge.net/

3) Install ffmpegsource2 (The contents go into the avisynth/plugins directory) http://code.google.com/p/ffmpegsource

4) Write the following in notepad in the same directory as your video, changing filenames to match. Save it and rename the extension from .txt to .avs

FFMpegSource2("sample.mpeg", atrack=-1)

5) Import the .avs file and conform (Interpret) the imported .avs as 1.333 PAR (.avs scripts are interpreted as 1:1 square pixel). The .avs is like a proxy that represents your footage, with pulldown removed

So it seems that this approach doesn't require any conversion/rendering of the captured footage, but rather uses AVSynth to reinterpret the footage within Premiere?

It look like the only drawback would be to insert file names into the .avs script if I have a gazillion clips to work with. I will look into the AVS batch scripters!

It would be waaay easier to use FCP if it interprets the footage correctly

I will try the AVS workflow, but my nightmares with FCP have been ongoing and severe. I've had occasional horror stories with Premiere CS3 as well, but they are nothing compared to my experience with FCP.

This is a known issue with that camera and similar canon cameras, a quick search:
http://www.dvxuser.com/V6/showthread.php?t=110397
http://www.dvxuser.com/V6/showthread.php?t=117029

Both of these threads seem to be primarily about footage down converted to SD in camera, whereas I'm editing HDV. But my understanding of pulldown removal and all that is admittedly minimal.

There maybe 3rd party utilities to remove the pulldown easily, like cineform

I thought about Cineform as a potential solution. Will look into it.


I will give the AVS workflow a try over the weekend! I just don't understand why I don't see people talk more about this issue with XHA1 footage and Premiere.

RPK

 

[PDR]

So it seems that this approach doesn't require any conversion/rendering of the captured footage, but rather uses AVSynth to reinterpret the footage within Premiere?

Correct, it's a frameserver, so to premiere it looks like footage with pulldown removed. Unfortunately , you will get red render bar and it will be slower due to the extra overhead

It look like the only drawback would be to insert file names into the .avs script if I have a gazillion clips to work with. I will look into the AVS batch scripters!

I think some users on the HV20 forums made some batch utilties, and some batch converters which do all the processing in 1 step to a lossless intermediate. You have to search to find them, I don't have the links handy. You could also just use the batch .avs files without transcoding (it's going to depend how fast your system is, and if it can handle the extra overhead)

Also, did you try other clips? It might be a problem with that particular clip (with the field repeats at the beginning that I mentioned earlier) . But I know for a fact that premiere has issues with 24p in 60i formats (from all Canon cameras, not necessarily just yours).

 

[RPK]

you will get red render bar and it will be slower due to the extra overhead

I assume I can render timeline previews just like before, or does the frameserver make rendering previews impossible?

batch converters which do all the processing in 1 step to a lossless intermediate

If conversion isn't necessary with the frameserver, wouldn't that be better in terms of maintaining image quality than actually converting the clips, even if it's to a near-lossless format?

Also, did you try other clips?

Yes. I've had the XHA1 since 2007 and have tried Premiere CS3 and CS4 every once in a while. Always with the same issue.

RPK

 

[PDR]

I assume I can render timeline previews just like before, or does the frameserver make rendering previews impossible?

Yes you can, but this gets tedious IMO. It would be nicer if Adobe folks got this fixed if you know what I mean....

If conversion isn't necessary with the frameserver, wouldn't that be better in terms of maintaining image quality than actually converting the clips, even if it's to a near-lossless format?

Yes, the frameserver is bit identical to the decoded frame. However, the current version of Premiere clamps the luma values of all RGB formats it doesn't natively understand to 16-235. This includes lossless formats like lagarith or huffyuv for example, and frameserved videos. There is an internal YUV=>RGB conversion and the values are clamped. e.g. If your camera has overshoots or highlights in this range, they will be clamped (instead of preserving 0-255). For most footage this won't matter, and you will won't see the difference. Premiere doesn't do this for uncompressed 10-bit 4:2:2 YUV footage. So if these overshoots are important to you (maybe for color correction), you could use that script to encode to uncompressed YUV footage to preserve that range

But examining your sample clip with a histogram or waveform monitor, there are no excursions beyond those "TV"levels 16-235. But that doesn't mean this is true for all samples from your camera

Interestingly, the latest version of Vegas seems to remove the pulldown on the footage correctly

 

[RPK]

OK, I couldn't resist and tried the AVS workflow. It's easy and the footage looks beautiful! Fantastic!

One strange thing I noticed: The AVS treated clip is 00:00:10:22 long vs. the original clip 00:00:10:21. When I place the original clip over the frameserver clip there is a noticeable drift.
I can only imagine that much longer clips will be significantly different in overall time. I wonder how this will affect my audio, especially when it's recorded separately and not part of the captured footage. I often sync up audio from captured footage with separately recorded audio, so the two need to match.

But the footage looks perfect!

RPK

 

[PDR]

Sometimes there are orphaned frames when doing the IVTC (or pulldown removal), but the embedded audio should be in sync. You try can sync external audio up manually , or with an avisynth script (you specify an audio delay with the frameserved audio). There is no way to use timecodes that I know of with this method

The other method (a bit harder) of frameserving gives a slightly different duration (1st method 10.844, 2nd method 10.969) these are in seconds.milliseconds, not seconds:frames as the usual Adobe notation (or 264 frames vs. 261 frames). DGIndex method usually gives very consistent results, but takes an extra step:

2nd method of IVTC:

1) Use DGIndex to index your .mpeg file. Leave video=>field operation to honor pulldown flags. File=>save project. This will generate a .d2v file and demux the audio (with the appropriate delay written in the audio name). A copy of DGDecode.dll has to be in the avisynth/plugins directory for it to autoload
http://neuron2.net/dgmpgdec/dgmpgdec.html

2) Download the TIVTC package, place the .dll in the avisynth/plugins folder
http://web.missouri.edu/~kes25c/TIVTCv105.zip

3) Open in notepad, do the same thing as before but use MPEG2Source

MPEG2Source("file.d2v")
TFM().TDecimate()

Try it out on longer clips and let me know how it goes. This only frameserves the video, not the audio. You can specify it to frameserve audio as well using AudioDub() , but if you are using external audio, it's just as easy to drop it on the timeline in Premiere

Cheers

 

[RPK]

I will try a long clip (45 minutes) over the weekend and will sync up external audio manually. I've always done it that way anyway and will report back if the external audio will stay in sync over the entire duration of the clip. If it doesn't I will try the DGindex pulldown removal.

I'm curious to see if using the frameserver will increase final render time when exporting. But if I render timeline preview files before anyway to make editing smoother I believe Premiere uses those for final export.

Can't wait to try this out!

 

[PDR]

It definitely takes longer. CPU is used for the pulldown removal. You will find method 1 faster than method 2 because TIVTC uses more CPU and is slower.

Before your final render, make sure you clear out your cache files (in the options), and make sure you do NOT use render preview files in AME (there is an option in the fly out menu)

Another method for smoother editing is to use those scripts to encode to a low quality intermediate as a proxy file, then switch to the final thing on final render. This way there is no CPU used for pulldown removal, and decoding is a breeze. The negatives are it takes a few extra steps.

 

[RPK]

I haven't had time yet to try out a long clip but would be glad to know why you advise to clear the cache and not use preview renders for export?

And I honestly never understood how to work with proxy files. Does Premiere provide an easy way to relink proxy clips to the full quality clips for final render?

RPK