HDLink vs Pr CS4 and more questions

[Roger Averdahl]

Despite my earlier rants regarding render times i have now bougth Neo HD because of the quality. :) (My opinions regarding render times remains the same though.)

I have a copule of questions regarding the quality of the files and the times it takes for Premiere Pro CS4 to generate Peak Files for the audio in the files:
1. Whats the difference in CFHD files converted by HDLink and the CFHD files converted with Adobe Media Encoder?:
a. The files converted by HDLink looks very different compared to the original AVCHD files while the time it takes for Premiere Pro CS4 to generate Peak Files when i import those files are normal.
b. The files converted by Adobe Media Encoder looks identical to the original AVCHD files while the times it takes for Premiere Pro to generate Peak Files are abnormal;

One hour of HDLink files takes approx 1.5 minutes while the same hour of files that Adobe Media Encoder converted takes approx 4.5 minutes to generate Peak Files for. (Right now i am doing a longer project and had to wait 40 minutes for the Peak Files to be created, thats why its an issue...)

2. Why does the files converted by HDLink look very different compared to the files they are converted from?

The image is overall darker than the original, green grass becomes noticable darker green and nice red colors looks oversaturated in all clips converted by HDLink. I see no difference between Low, Medium, High, Filmscan and Filmscan 2 other than the file sizes.

3. Are the files created by HDLink 8-bit, 10-bit or 32-bit? I read in this forum about "we use 32-bit float..." while the specifications only mention 10-bit I/O Precision while the files HDLink creates loks more like 8-bit files than 10-bit. What gives? :)

4. In HDLinks Preferences the default option when converting/capturing 1080i is to deinterlace the files. Why do one ever want to deinterlace files upon conversion/capturing by default? That deinterlacing enhances your footage is one of the biggest myth in the video world. Deinterlacing should only be done when necessary, not just because it can be done and especially not be done by default because it degrades the quality of the video. Or, did i miss something? :)

5. Where can i find a manual for HDLink that are up to date? For example, the one that comes with Neo HD dont mention anything about Concatenate Chunks.

[Robert Young]

Originally Posted by Roger Averdahl

Whats the difference in CFHD files converted by HDLink and the CFHD files converted with Adobe Media Encoder?:
a. The files converted by HDLink looks very different compared to the original AVCHD files
2. Why does the files converted by HDLink look very different compared to the files they are converted from?

The image is overall darker than the original, green grass becomes noticable darker green and nice red colors looks oversaturated in all clips converted by HDLink.

Good questions.
I was using PHD 4/ CS4 to downscale 1920x10870 60i CFHD to 720x480 60i UFF CF.avi for compression to DVD:
When doing the downscale with HDLink, the resulting SD images were very soft and the color was noticably shifted from the original
When doing the downscale from CS4 (AME) with the CF compressor, the resulting SD image colors were good, but there was unacceptable interlace artifact (combing).
I ended up converting the CFHD 1920 60i directly to m2v with AME- this was the best result, but not as good as I used to get with PHD3.x/ CS3 to make CFSD master.avi, then m2v conversion.
Is this difference between PHD3/CS4 and PHD4/CS4 related to the lack of real time CF function & presets??
My current work around is to transfer my CFHD.avi masters to my old CS3 machine to get the best DVD output.
Any better solution??

[David Newman]

Roger,

1. Likely audio an audio interleave difference, video compression is the same. More disk seeks will stake longer to import.

2. Adobe is very likely wrong, which it is with most of the native formats (HDV for sure, and likely AVCHD as well.) In HDLink's AVCHD transcodes, we do everything in the native YUV colorspace, nothing is tansformed through RGB, no Rec709 or Rec601 issues (which plagues Premiere.) Premiere YUV processing is still inherently in 601 colorspace, which not the colorspace for all the new sources. If you have a camera that generates color bars you can see very quickly so see that Premiere is wrong for native decodes. Red comes out more orange and the greens are wrong.

3. All 4:2:2 files are 10-bit internally. Premiere decodes are either 8-bit or 32-bit, depending on user request.

4. We don't default to deinterlace, it is just saving an earlier setting.

5. Concatenate Chunks is for Grass Valley Infinite cameras only.


Robert,

The HD and SD color conversion is accurate, it is used for high-end facillities DVD finishing. Don't use Premiere to test that is at get the colorspace wrong too often, same goes for the graphics card for Windows Media playback. Once the downscale is completed, compare the results with a tool like VirtualDub. 1080i60 to 480i60 will always be soft. I always deinterlace HD to better SD results.

[Robert Young]

Originally Posted by David Newman

Robert,
The HD and SD color conversion is accurate, it is used for high-end facillities DVD finishing. Don't use Premiere to test that is at get the colorspace wrong too often, same goes for the graphics card for Windows Media playback. Once the downscale is completed, compare the results with a tool like VirtualDub. 1080i60 to 480i60 will always be soft. I always deinterlace HD to better SD results.

What I am noticing is that my successful CS3/PHD3.x DVD workflow: CFHD60i> CFSD60i> m2v60i, which has produced many stunning, artifact free DVDs, is not giving me the same results within CS4/PHD4.0.9. If I transfer the CFHD master to the CS3/PHD3 machine, downscale & transcode to m2v it's perfect. I'm trying to figure out what's the difference.
As for colorspace: When I downscale the CFHD in HDLink (PHD4.0.9) I get quite different colors than when I downscale the same footage with the CF engine from within CS4 AME That is not the case with CS3/PHD3.x
It seems that something is quite different in CS4/PHD4.
P.S. The color from the CS4/CF/AME downscale matches the BluRay version colors as seen on HDTV, whereas the HDLink (PHD4) downscale does not.

[Roger Averdahl]

Originally Posted by David Newman

1. Likely audio an audio interleave difference, video compression is the same. More disk seeks will stake longer to import.

O.k, but why does it take soooo long time with a CFHD file from Pr when lets say Blackmagic Design uncompressed 10-bit @ 1920 x 1080 conforms the audio waaaay faster. The file size in the Blackmagic Design file is approx +6GB/minute so the disk seeks are much much more intensive on a single SATA drive than it is with a CFHD file. Both files have the same length.

One hour HDLink footage and One hour CFHD made by Premiere Pro has the same file size, so there must be something else than disk seeks alone. Could this be the same issue that make the render times from CFHD > CFHD take ages, or at least related to it?

Originally Posted by David Newman

2. Adobe is very likely wrong, which it is with most of the native formats (HDV for sure, and likely AVCHD as well.)

That bug was fixed in CS4. If CFHD is CFHD no matter who created it, HDLink or Pr, there must be something else.

Or, if they are likely wrong, i hope CF not stops there and not trying to find the bug because it may be someone elses fault?

I will file TT's for this so it is logged. :)

Originally Posted by David Newman

4. We don't default to deinterlace, it is just saving an earlier setting.

Hmm, i never deinterlace and have thus not changed HDLink to do that. I will retry it on a newly installed computer. Or, how do i manually delete HDLink's Preferences so i can test it on a computer with Neo HD already installed? :)

Originally Posted by David Newman

5. Concatenate Chunks is for Grass Valley Infinite cameras only.

O.k, thanks for the answer! :) But when will an updated manual to HDLink be aviable?

[David Newman]

1. May answer here is still correct, the size of the seek is not as relavant as the number of seeks. It is the audio interleave -- Premiere output to CFHD uses a frame interleave and HDLink uses a one second interleave (faster.) As Premiere's AVCHD decode is wrong, I would be using HDLink to generate all your sources, as it batches, has the correct color, and is faster.

2. We can not be responsible for all Adobe bugs, although we try to address as many as we can. We can only do our best to make sure HDLink decodes camera source correctly for YUV based accurate transcodes (currently I believe this is spot on.) Adobe does their own avchd decodes, if they get the color space wrong we can't control that.

4. Turn deinterlace off and it stays off, is that not working for you? To clear the settings, delete registry keys in HKLM/Software/AspectHDCapture

5. Manual will always behind the product, we add features too quickly. When they is a lull in new cameras and modes we will get to polishing the manuals. One reason we offer so much online support, it will always be more current.

[Jim Snow]

Originally Posted by David Newman

I always deinterlace HD to better SD results.

How do you deal with edge blurring on motion due to temporal shift between the fields?

[David Newman]

Deinterlace in Premiere only uses one field, so 1920x1080 60i, becomes 1920x540 30p, before it is scaled to the target 720x480. No edge issues. Or HDLink uses both fields, producing an effective 1920x720 30p before scaling to the target. First has no added motion blur, second is cleaner and smooth, blur with more motion blur. I use the second method, as I typically have a more stationary camera, the blur is not a big factor.

[Jim Snow]

Originally Posted by David Newman

Deinterlace in Premiere only uses one field, so 1920x1080 60i, becomes 1920x540 30p, before it is scaled to the target 720x480. No edge issues. Or HDLink uses both fields, producing an effective 1920x720 30p before scaling to the target. First has no added motion blur, second is cleaner and smooth, blur with more motion blur. I use the second method, as I typically have a more stationary camera, the blur is not a big factor.

Does Vegas Pro 8 also use one field?

[David Newman]

Test and find out, but I expect so.

[John Rich]

Roger and Bob,

I know this is off topic to some extent, but I was wondering if you had ever tried Dan Isaacs's technique of down scaling to SD and compared it to what Cineform can accomplish?
This is sort of a summary link of his procedures, in case you don't have it already.
http://www.precomposed.com/blog/2009...-best-methods/

Personally, I've always been satisfied with the PremPro CS3 and Prospect HD procedure that Bob has outlined several times in the past, but if such a comparison has been done, it would be great to know the results of it, particularly since Prem Pro CS4 apparently has been giving some problems.
Thanks,
John Rich

[Robert Young]

Originally Posted by John Rich

Roger and Bob,

I know this is off topic to some extent, but I was wondering if you had ever tried Dan Isaacs's technique of down scaling to SD and compared it to what Cineform can accomplish?

I had tried this workflow a bit (debug frameserver/ AVI Synth) back when Dan Isaac was developing it- but am no expert. There are many who swear by it.
I have been so consistantly satisfied with the CF workflow that I haven't pursued other methods for a while. Because of the hang ups with CS4/ PHD4, I just retested my CFHD>DVD procedure.
Original footage was shot by client as HDV on tape with high end Sony consumer cam.
Footage converted to CFHD 1920x1080 60i on capture. Editing done in CS4.
All DVDs were viewed with upscaling DVD player, HDMI out to 50" HDTV.
All DVD encoding was 1 Pass 6mbs VBR with no filtering or tweaking in Procoder 3

1) Within CS4, Exported CFHD 1920 60i timeline directly to m2v with AME- DVD images: a little soft, color good, motion rendering good, no combing, occasional slight line twitter.

Then I transferred the CFHD 1920 60i edited master from CS4 to the CS3 machine. This is an event program with lots of motion, panning, etc.
2) Deinterlace CFHD master to 1920 30p in PHDv3 HDLink> m2v with Procoder 3> DVD- images show good color, poor motion rendering, very soft image with motion, no interlace artifacts.
3) Placed the CFHD 1920 60i master on a CS3 timeline> Export to CF 480x720 60i UFF SD master> compress the SD master to m2v with Procoder 3> DVD- images on HDTV are crisp (but obviously not quite HD), good color, excellent motion rendering, and no artifact.
Conclusion- workflow #3) still provides the best quality DVDs I have ever been able to produce. IMO in the same ballpark as hardware encoded commercial DVDs
I am currently not able to reproduce this workflow with PHD4/CS4, so for now will continue to transfer the HD master to CS3 system for DVD output.