4k downscaled to 1080p

Hey everybody,

Im about to change my camera, and im seriously considering buying one that can shoot in 4k to downscale it to 1080p to get more details and space colors.

Ive seen on many forums and websites, that this process can change the color space from 4:2:0 to 4:4:4 since 4 pixels becomes 1 pixel but it contains 4x more informations, which can be a huge upgrade for chroma key footage (which is my case).

Does anybody else heard about it ? Is it reliable ? Im very curious to have your opinions about it !

«1

Comments

  • Triem23
    Triem23 Posts: 20,080 Ambassador

    In short, yes, it works. However, you don't just throw your footage on the timeline, scale it, and magically have 4:4:4 colorspace. You have to transcode your footage using software that supports bi-cubic sampling. For Windows you need to go to DNxHD, for Mac, ProRes. 

    Now, you'll have to specifically ssetup your transcode options for this, but, be warned, you'll end up with huge file sizes. You'll have to use a 10-bit 444 output, and you're looking at a data rate of over 350Mb/sec. Roughly a gigabyte for every 25 seconds of footage. 

  • foleyproduction
    foleyproduction Posts: 110
    edited July 2015

    Thanx for the quick reply, so basically i transcode the footage to dnxhd, then i import it in davinci resolve lite in 4k and render it into 1080p ?

  • Triem23
    Triem23 Posts: 20,080 Ambassador

    I don't use Davinci, but if you can save out of Davinci in DNxHD, you could take your footage directly into Davinci and just scale and render and save yourself an intermediate transcode. 

    Otherwise, I think what you described would work, but maybe a Davinci user could give a definitive answer. 

  • Andy001z
    Andy001z Posts: 3,152 Ambassador

    My word, working in 4K means some big files.... this must suck on low spec PCs.

  • foleyproduction
    foleyproduction Posts: 110
    edited July 2015

    Ive just checked and in the render section, there is a DNXHD 1080p 444 codec with MOV format, so i think what you suggest is possible, import my original footage (AVCHD 4K) in davinci to export it in DNXHD1080p 444.

    Hitfilm will not have any problem to run the DNXHD codec in MOV ?

  • Just to be more precise and if other people are interested in this process, i found that link with a confirmation from the CEO of cineform, i just don't get the last part where he says "As long as you up convert to RGB and 10/16 bit first ..." CEO from cineform confirms 4k to 1080p 444

  • Triem23
    Triem23 Posts: 20,080 Ambassador
    edited July 2015

    ... Which version of Hitfilm are you running?

    Excuse me while I use arcane symbols to summon staff: Hey, @Ady there have been past issues with 4GB+ of files before. You guys were trying to work around the Apple-caused QuickTime issues that were causing Hitfilm to stall. Gotta guess to how a 10-bit, 352mpbs, 4:4:4 colorsampling DNxHD codec in MOV wrapper would perform? That's 4Gb of file in 2 minutes of footage! 

  • im running the last version HITFILM PRO 3

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast
    edited July 2015

    I would try importing the original footage into a Resolve 16 bit project, scale and export using Cineform 4:4:4: Film 1 or 2 instead of DNxHD because Cineform can be used with an AVI wrapper avoiding the QuickTime.MOV issues completely.

  • AxelWilkinson
    AxelWilkinson Posts: 5,252 Staff

    @Triem, HitFilm 3 has much improved Quicktime handling, which should eliminate those RAM limitation issues. That being said, that sort of datarate is going to want a very capable system in order to keep up. ;)

  • Andy001z
    Andy001z Posts: 3,152 Ambassador

    @AxelWilkinson Is this going to be one of the drawbacks of accessible 4K footage, please will expect to be able to view videos in 4K, but in order to do anything funky with those 4K videos you are going to need more and more power. I suppose my point is I think computing power caught up with HD video and was pretty OK at managing it on a standard set of hardware, but now we are starting to push the boundaries again. This thread has certainly made me think hard about any future camera specs to ensure they can record in lower resolutions as well as 4K. 

  • Triem23
    Triem23 Posts: 20,080 Ambassador

    @Andy001z that's where we were when digital video started, and what we went back to at the dawn of HD. The thing about 4K is, you really don't need to output 4K yet... You can't tell the  difference on anything under, like a 50/60 inch screen between 1080 and 4k. It's major advantages right now are in giving room to crop, and in "creating" 4:4:4 1080 footage, otherwise, downsampling 4k to 1080 can help reduce compression artefacts and result in sharp 1080 footage. 

  • foleyproduction
    foleyproduction Posts: 110
    edited July 2015

    @Triem23 yeah this is the reason why i want to upgrade to 4k, getting the best 1080p version possible with as many datas as possible and fewest noise, which is crucial for chroma key when you don't have thousands of euros to spend in a 444 camera ...

    After few researches it seems that just rendering out 4k to 1080p or import 4k footage to a 1080p timeline improve the quality, but to get the 444 color space looks more complicated, english is not my native langage so sometimes i get lost with technical explanations ...

    Id be glad to talk about chroma key techniques with guys on this forum, is there many people doing it here ?

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast

    The key detail in a 4K 4:2:0 to 1080 4:4:4 conversion is the scaling needs to be done in an RGB form. If a tool scales the 4:2:0 UV chroma component intact, then you cannot get the 4:4:4 1080 quality result.

    I'll bet most tools do not preserve the chroma channel subsampling intact through the edit data stream so the trick probably works with any tool.

  • @NormanPCN could you please give a workflow as an example because im a bit lost when it comes to that RGB form, the CEO from cineform also explained it in the link ive published in a previous message.

     

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast

    @foleyproduction What NormanPCN is saying is a very valid point but as he also points out probably a moot one as you would have to go out of your way to find a tool that doesn't  work in an RGB color space

    Let's say you have a 4k clip and the color space is stored as 4:2:0 YUV. When you import that clip into just about anything like HitFilm, Vegas, Final Cut, Resolve etc the color space gets converted to an RGB color space. Normally this RGB color space is limited to 8 bits per channel or RGB24 which gives you roughly 16.7 million possible colors. If you scale down in this color space you won't get the benefits you're looking for and this is what the article you linked to is talking about. In order to get the benefits you're after you not only need to be working in an RGB color space but it also needs to have a higher bit depth of at least 10 bits per channel. For most applications this would be an RGB32 color space but some like Resolve also support 16 bits per channel RGB48.

    So a sample workflow would be to start a DaVinci Resolve project with a bit depth of 16, import your 4k footage to scale it then export to Cineform 4:4:4. (Or DNx). Setting the project to 16 bits per channel is what will "up convert" your footage to RGB 16 bits per channel just like the Cineform CEO described.

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast

    @foleyproduction, you ask about workflow. There really is no workflow you can adopt for this 4K>-1080 4:4:4 thing to work. Either the application doing the scaling works on an intermediate RGB or YUV form without chroma subsampling or it does not. Nothing you can do about it.

    So really the workflow is simply to just transcode the 4K 4:2:0 media to a 1080 4:4:4 format.

    As I stated I suspect that most tools these days operate internally in RGB or YUV format without chroma subsampling even when the source had subsampling. It just makes sense to operate this way.

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast

    @NormanPCN From what I understand simple transcdoing won't work for this process. If you start with 4k 4:2:0 8 bits per channel then a simple transcode to 1080 4:4:4 isn't going to gain you anything because you aren't really storing any new or extra information. All you're really doing is telling the codec not to sub sample the color information you're feeding it and you're feeding it color data from just the pixels that are being kept.

    You have to "up convert" not just to an RGB color space but also a higher bit depth color space, then scale and export. When you do this then the color information doesn't come from just the pixels you are keeping like in a straight transcode but also all the neighboring pixels you're getting rid of. This "extra" information from the discarded pixels then gets appended to the remaining pixels as increased bit depth information,  

    In order to store all that information you need to use a codec that won't sub sample and that can handle more than 8 bits per channel so not just a 4:4:4 codec but also one with 10 or 16 bits per channel bit depth.

    So workflow wise the first thing you need is something works in an RGB color space. This is almost everything but note that Adobe Premiere is an exception which I'll get to later. The second thing is it also has to support a bit depth greater than 8 bits per channel. This is quite a few things but you have to be sure to specify the greater bit depth because the default will generally be 8. The third thing you need is an export format that supports the greater bit depth and is compatible with whatever else you intend on using. For a 10 bit workflow that means Avid DNx, Cineform, Grass Valley HQX or ProRes. OpenEXR and DPX sequences support 16 bits per channel workflows. Out of all of those Avid DNx and Cineform will work across the broadest range of hardware and software. Right now the best free tool that's going to provide the desired results is probably DaVinci Resolve. Start a project set to RGB and a bit depth of 16, import the 4k footage, put it on a timeline then export to 1080 using DNxHD or Cineform 4:4:4  @ 10 bits per channel. An After Effects 16 bit comp can produce the desired result too if you're already an Adobe user.

    Special note on Adobe Premiere - Premiere will actually maintain YUV color throughout the entire pipeline if at all possible. There is a conversion to RGB for playback but this is done on the fly and does not affect the render pipeline. So if you import a YUV clip, don't do anything to it and then export it to a YUV codec there will not be any conversion to RGB. Conversion to RGB will only happen when there's an operation that requires it like adding an effect that only works in RGB. When conversion is required Premiere converts to RGB32 float.

  • __simon__
    __simon__ Posts: 113 Just Starting Out

    All this talk of transcoding and other apps. What's wrong with simply using HitFilm?

    Presuming the end goal is to do some VFX in HitFilm then why not;

    1. Project Settings: Color Bit Depth: 16 bit Float
    2. Create 1920x1080 Composite
    3. Stick your UHD footage into Comp and Scale 50%
    4. Right click on Comp and "Make Proxy"

    Of course on a laptop proxies are probably a no go due to limited hard drive space.

  • Triem23
    Triem23 Posts: 20,080 Ambassador

    My main objection to proxies is just that proxies live in their own little subfolder where Hitfilm expects to find them (unless one can load a proxy as a media file?), so, when I do my project archive, I have to track down the correct folder, extract the proxy and add it to the project backup, then, if I reload this project back from a backdrive drive to an active work drive I have to find those proxies and stick 'em back in the correct Hitfilm subfolder--unless I skip that whole thing, in which case, if I return to the project at a later date, I have to take the time to re-render all the proxies again.

    What you describe @_simon_ totally works, but that's MY reason for not going that route.

    Yes, I archive all my old project files, and, yes, I do have to sometimes go back to an old project that might be ten years old.

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast

    @__simon__  That totally works but it isn't very portable across different applications and the export options that would preserve everything are limited too. 

  • Palacono
    Palacono Posts: 3,423 Enthusiast

    @Triem23 does it work though? I tried @__simon__'s suggestion,  but can't tell if it works inside Hitfilm.

    While you can set the rendering option in the Project settings to 16 bit float,  after that you have no way to tell how Hitfilm uses that setting. You can't change it at all in HF2E, so no idea what it's using there.

    Checking the Layer properties in the Composite timeline says "24 bit", so 8x3 presumably (why a different format from it saying 8 and 16 elsewhere?) and while in the rendering options for AVI and OPENEXR there are RGB and RGBA colour depths, but I don't know how this affects what's come before it or how/if the file is transcoded from the source properties to the project properties and (possibly again) to the output properties.

    There is no field in the Composite shot properties that tells you what Bit Depth it is using at all - if this is where the conversion is happening when you make a Proxy file from an 8 bit source to a 16bit intermediate. And why is that the best way, rather than rendering anyway?

    How do I know that when rendering  .MP4 files (as I don't have the option to output Cineform in AVI) that Hitfilm has actually taken the 4:2:0 4k file, converted it to 16bit, and combined it into an approximation of either 4:4:4 or 4:2:2 before transcoding it down 50% to 1080p for export correctly? Where's the evidence (other than doing it twice, once with a 16bit float project setting and one with 8 bit and looking at the two video files?)

  • Triem23
    Triem23 Posts: 20,080 Ambassador
    edited August 2015

    HF Express is 8-bit only. 

    Yes, the composite timeline says 24-bit,which is 8x3. The difference here is that the bit-depth is assigning either 8 or 16 bits to each color channel, where the overall timeline is showing total bit-depth. When you set Hitfilm to 16-bit float it internally processes everything in 16-bits/channel then renders out to the viewer in 8 bits/channel, but I could be wrong. 

    I suspect that Hitfilm's mp4 export is 4:2:0. I don't recall seeing options for the High444pp option. I think to get the export out of Hitfilm Pro you have to export 16-bit Open EXR and still have to transcode. 

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast

    HF AVC/H.264, aka MP4, export is 4:2:0. MediaInfo (free) is your best friend when figuring out what is video file contains.

    Hitfilm should add an XAVC-Intra output option. It is typically 10-bit 4:2:2. They already license/use the Mainconcept encoders so it is trivial to switch/use the high-bit 4:2:2 broadcast encoder.

    XAVC-Intra is just AVC which all editors natively support. XAVC is just a name. Any AVC Intra only 4:2:2 output works and you don't need to fiddle with DNxHD or Prores or Cineform which are not natively supported across the major editors.

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited August 2015

    OK, @Triem23 ,@Alladin4d , @NormanPCN  and @Ady, to answer my own question: I took some 4K footage, transformed it down 50% and exported it as a 1080p MP4 after setting up an 8 bit project.

    Result: like it had simply thrown away 3/4 of the pixels without looking at them. Nice and sharp, but lots of aliasing on contrasting lines (buildings).

    Did the same again as a 16 bit project and the result was much smoother, almost no aliasing, but still sharp, so looks like those 3/4 of pixels it threw away were actually used for something after all.

    Just to test further, I exported an 8 bit project again at 51% and 49% reductions to see if some blending would happen there as it wasn't a simple skip-every-2nd-pixel, but...nope. Still lots of aliasing. One more time at 50% after putting a very slight blur on it first (2px, 1 iteration) and still same noisy result.

    Could have blurred it more, perhaps, but instead:  my new default setting for all my projects is now: 16 bit. Result!

    I wonder if this might improve the horrible aliasing results I get with text, which is so unusable when it is moving that I no longer bother with it and instead use a PNG with what I want written on it, scaled down a little and dropped where I need it. Scaling down actual larger text never worked as a solution, as it was apparently simply rendered at the smaller size, rather than drawn large, then reduced, which would have produced some nicer anti-aliased edges. Instead, you get hard, crawling edges. Probably the right solution for scaling up, but not so good for scaling it down.

    And yes, would be nice if Hitilm exported in more formats. We've got the codecs installed, why won't it use them?

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast
    • We've got the codecs installed, why won't it use them?

    If you are on a Mac then one can there there is such a thing as a system installed  codec since Quicktime effectively is the system and a Quicktime codec is thus effectively pervasive.

    Not so on Windows. We have

    Video for Windows. Ancient and depreciated since the early 90's. AVI files only.

    DirectShow. The VfW replacement. I know of no editor that uses this.

    Media Foundation. Windows 7+. Same as DirectShow for usage.

    Installing a DirectShow codec does nothing for VfW and so on. Most codec packs out there install DirectShow codecs, which video players use to playback anything and everything.

    Hitfilm on Windows has some AVI output but is very restricted so maybe it is not actually using the VfW system during render but writing AVI directly for those restricted codecs.

    Hitfilm does seem to use VfW/AVI and Quicktime/MOV generically to read anything that has a codec installed into those respective subsystems.

  • Wow im really glad to see all this replies and propositions, concerning the solution described by @Simon looks very efficient and quick, but i wonder if compositing directly on a 4k footage would not be too heavy and slow down the computer (i have a I5 processor quadcore with 8GB and GPU 1GB NVIDIA) ...

    It seems that @aladdin4d's "workflow" in davinci could be more accurate to get the best results

    Soon, i will buy the LX100 from Panasonic, hopefully i will do tests and give feedbacks here.

     

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast

    @foleyproduction - The short version is the conversion to RGB triggers a complete recalculation of the color information and the extra bit depth allows you to take advantage of color information scavenged from discarded pixels. Resolve is one way to get there and After Effects will supposedly get you there too. I'm sure there are other ways but Resolve is very effective and very free.

  • monica66
    monica66 Posts: 22

    Here is a tutorial I found on Google which should be helpful to you.

  • Ubi
    Ubi Posts: 85

    Really interesting thread!

    I'm currently editing 4K cineform transcoded gopro footage (resized to 50%) directly into HF3 and it's enough smooth for me. When i switch to 16 bit process in project the machine crawls so i guess it's ok to do all the editing in 8 bit and switch to 16 bit just before exporting?