[Kati Haapamäki]

The story continues…

So I didn’t find a proper tool for HD to SD downconversion. I went deep and started to study more. I ended up writing own plugin and went even deeper… But now I am back and I am back with a brand new plugin for After Effects 🙂

I want to believe this plugin beats many commercial software scalers. But you can find out for yourself, because I am giving it away FOR FREE! 😉

Plugin can be downloaded from my site https://sites.google.com/site/katisvideotoolbox/smooth-bspline-scaler. There’s some test images and other information about it.

It’s Pixel Bender code and works only with CS5+ on both Mac and Windows (There’s is tiny possibility that it could work also in CS4 but I have not tested).

Kati Haapamäki

Ps. Because this is FCP X forum, I must remind that FCP X does very decent downscaling. There’s no need of doing round trip to AE. Upscaling may be a different case tho..

[Kati Haapamäki]

Well.. I installed Compressor 4. Results looked quite promising when I started to play with anti-alias slider. But then I remembered why I have never liked Compressor. It is buggy and so is its newest version.

It creates quite a lot corrupted (bright colored) pixels, sometimes in rather big area. It’s totally unacceptable. Compressor 4 is not a solution. Sniff..

Kati

[Kati Haapamäki]

We have been struggling with the same problem for months. First with LHe and then with LHi.

Then I had an idea to try to do things at a lower pace. With a little more delay between moving the playhead and pressing play button did the trick. I couldn’t get audio out of sync with this methdod, and asked our editor to try this too.

Our editor has reported that he has had no sync issues since. He is just too fast editor for our Kona cards 🙂

But good to know that disabling audio scrubbing may help too.

Kati Haapamäki
System Manager
MTV Media
On-Air Promotions

[Kati Haapamäki]

Hi Drea.

My wrote my answers beneath your questions.

a. I assume this means the downscaling is done of just the raw HD footage (1920×1080 30p) vs. the final edited material? (I’ll be adding titles, transitions, and overlaying HD clips from 2 camera angles). I assume you are recommending adjusting the raw footage (I will also be sparingly blending in a few clips shot in SD so it would make sense to add this after the downsample).

You add all the SD stuff afterwards, so it stays intact. It’s also wise to make all the graphics in the SD size and add them after downsample.

b. What are the actual steps to take for downscaling in FCP? So far, I’ve dragged and dropped HD clips into an SD timeline – is that in fact the downscaling method recommended or is there another way? Do I create an HD timeline for all of my HD footage, add the blur, and then what?

When you throw your HD clips into SD timeline, FCP resizes them using it’s own not-so-great downsample algorithm. You can nest your HD timeline, and add blur in to nested clip and then edit the whole HD sequence into SD timeline. (FCP scales the HD sequence automatically). Then you can add the sharpening. And edit you SD stuff to the that sequence.

But to get this working you must have all your footage and all your sequences have set their field dominance to None. Also you have to make sure that Motion Filtering Quality is set to “Normal” on you sequence settings. Other values seems to produce pure crap.

If you use Unsharp Mask, uncheck Luminosity checkbox. It seems to produce banding.

c. Your answer to b. should then reveal when/where I add the unsharp mask (your recommendations: add a little amount of Unsharp Mask afterwards to get some sharpness back. Try amount of 20 to 30 and radius of 1.0.)

d. And if I follow the above recommendations of blur/downscale/unsharp mask (hopefully with further clarification from you), will the results be markedly better than if I just ran with dropping the raw HD clips into an SD timeline and editing away and exporting as QT movie and on to DVD? I’m trying to judge how bad the latter route would be given my audience’s expectations and how little time I have to complete the project (of course!)

Markedly better? I bet most of the people are so blind that they can’t tell the difference nor they don’t care about quality. You have to decide by yourself if it’s worth of it. Make some test and use properly adjusted television or video monitor to compare results.

But this is a kind of all around fix and very much a compromise. It depends a lot on your original material weather this helps or not. If your footage is already soft, there’s probably not much use of this. But if you have problems with too sharp jagged edges and flickering this may help. Don’t take my filter values granted. Play with them and find out what’s best for you. You may want more post-sharpening for example.

There sure are better methods around, but if you’re stuck with FCP 6, you can try this. Not never tested this with FCP 6 tho, but i don’t think there’s any difference between versions 6 and 7.

Can’t wait getting my hands on Compressor 4….

Kati

[Kati Haapamäki]

Thanks for the tip Andy.

I have completely forgotten the existence of Compressor 4. I never liked Compressor, 3 but I really see hope in version 4. If it only has as high quality image processing as FCP X has built in. Gotta try it tomorrow.

Kati

[Kati Haapamäki]

I think that it’s a common pitfall to judge different downscaling methods only by looking at image sharpness. The problem with many software scalers is, that they produce *too* sharp results. Sharpness usually means also aliased (=jagged) edges, moire and flicker in thin, high contrast lines.

When you do downsampling, the original image must be low-pass filtered, for example by pre-blurring. If it’s not, the result is sharp, but also aliased with ugly edges. Of course this depends on the original material and how sharp it is, but generally low-pass pre-filtering is mandatory. Besides the bicubic algorithm, that many software scalers use, is not good with lower than 0.5 scaling factor. 

I have been doing some testing with After Effects, Premiere Pro, Final Cut Pro 7 / X, Mpeg StreamClip and AJA Kona hardware scaling. The best so far has been FCP X. It produces very nice anti-aliased image. It also seems to do the processing in linear 1.0 gamma, which gives generally better results. Now I hear some arguments that it’s result is too soft. Yes, it’s a bit soft, but remember it’s SD, not HD. It’s not supposed to be sharp 🙂 Besides FCP X performs a correct color space conversion between HD and SD, which indeed is quite exceptional. Anyway if you think the result is too soft for you, try to add some sharpening afterwards.

As a huge fan of After Effects, I must have faced the fact that doesn’t do a good job with downscaling. Its bicubic resampling really needs some pre-blurring. But you can help yourself. Try to add some Gaussian Blur before scaling with amount of 1.2 to 1.5, if you are converting from HD1080 to SD, and less if you’re converting from HD720. And again, you can add a little amount of Unsharp Mask afterwards to get some sharpness back. Try amount of 20 to 30 and radius of 1.0. 

Premiere Pro is not doing much better than AE without help of CUDA. I just read that with CUDA, PPro uses lanczos2 algorithm instead of bicubic, but I haven’t had possibility to test it. I’ll keep you posted, if I get some tests done with CUDA.

FCP 7 has basically the same problem in progressive scaling than AE. You can use the same pre-blurring/post-sharpening method than with AE. Results are very decent, if you can use progressive workflow (footage and sequences all set to “none fields”). If you have interlaced material, it’s not that happy story. The result is soft even without any pre-blurring and sharp edges looks weird and ugly like they were duplicated or something. FCP 7 has some severe problems with handling interlaced material. But I think you all knew this already.

Mpeg StreamClip is not a solution for HD to SD conversion. It’s downscaling is way too overrated. It falls into the same category than other of those non-blurring methods, being even worse than AE. But the worst part is that it doesn’t do YCbCr coding conversion between HD and SD. HD and SD formats use different RGB <–> YCbCr translation matrices and ignoring this step gives false colors. I am not talking about the lack of color space conversion here. This is a lot bigger issue. But notice this problem only applies to codecs that are using YCbCr color values (like most video codecs), not RGB.

I have also been trying to find a proper workflow to do HD to SD conversion. Although I can get good results with FCP X or AE+pre-blurring, neither of them is a fluent workflow for me. I am looking for a stand-alone batch converter that can handle the task without any extra pain. If you just know any one that does decent downscaling, please share your knowledge…

—————————-
Kati Haapamäki
System Manager
MTV Media
On-Air Promotions