Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
420, 422, 444, rescaling and colors flame
  • 114 Replies sorted by
  • I'm sorry, I can't seem to find anything about bits in that link?

    Dithering in the same way that a printer dithers to make different shades of gray on a piece of paper. If the 4k 8-bit image is properly dithered, the downscaling will lead to a smooth gradation equivalent of 10-bit. However, since dithering can't capture the exact gradations of a source image (the raw image before data is thrown away) the resulting 10-bit gradation will be a mathematical guesswork - similiar to the way Red MX guesses at colors based on a bayer pattern.

  • If the 4k 8-bit image is properly dithered, the downscaling will lead to a smooth gradation equivalent of 10-bit.

    Logic :-) As it is used in exactly reverse situation were you want to use limited values to represent something close to original. One of extreme cases is representing grayscale image using only b/w :-)

    If you look at thin in reverse - converting high res dithered b/w image into grayscale (and read my link, yes link does not have anything about colors, but it explains why DR range in bigger after rescale) you'll understand were additional bits come from.

  • I think we're on the same page, hehe :) I just wanted to point out that it will be an inferior 10-bit compared to one diverged from an actual 10-bit source, the same way a bayer pattern yields inferior color compared to real RGB stripes. Thus the 10-bit 4:4:4 from downscaled 4k isn't necessarily better than a "real" 1080p 10-bit 4:2:2, like people seem to imply in relation to the GH4

  • I just wanted to point out that it will be an inferior 10-bit compared to one diverged from an actual 10-bit source.

    If you'll had source 4K 10bit in AVC file it could be cool. But. Most of this new bits will be quite noisy. Yes, in ideal noiseless sensor it is good advantage, not so much in 4K on m43 sensors.

  • @Mistas Yes, good way of thinking about it. Just keep in mind that the 4 approximated pixels don't necessarily equate to a "precise" pixel in terms of color they hit. The actual number of bits will in practice, depending on source material, vary between 8bits and 10bits somewhere, but experts seem to suggest it is closer to 10 than 8.

  • @Vitaliy_Kiselev, with all you logic, please let me know how can you extract the 10bit dynamic range from a 8bit source? I mean "push" dynamic range in post, I mean get juice from the clipped highlight and the black darkness. Please don't tell me to "light the scene properly" or "get the right exposure", we know that even the right exposure have its limits, and some light are not 100% controlled (sun for example). Please, allow me to understand how these extreme dark "blacks" or clipped "whites" can be pushed, if they have only 8bit depth? From where this retrieved data will come from, if they nearby pixels from the 4k source are also 8bit darkness or 8bit clipped white? With 10bit data I can retrieve some over or under exposure parts because the elastic dynamic range of the 10bit. Let me really know how can you do that from 8bit? I already read all your recommend links and have found nothing about it, please let me know how to do it.

  • @ze-cahue, Don't confuse bit-depth with dynamic range. It is separate. You can have a camera with 8-bit and high dynamic range, or a 10-bit camera with low dynamic range. Pf course you would ideally get both.

  • So tell me if I understand this right. 8bit 4k downscaled to 1080 yields a 444 image with "virtual" (for lack of a better term) 10 bit? This virtual 10 bit is good but can be expected to be noisier than native 10 bit?

  • @brianl, the exact result will depend on the algorithm that Panasonic use when they convert 10 to 8 bit. An ideal implementation would actually make sure 4 pixels combined contain the actual bit representation of a single pixel in 1080p 10bit. I.e it COULD be true 10bit to recover. However, not sure if Panasonic chose to spend the processing power..

  • with all you logic, please let me know how can you extract the 10bit dynamic range from a 8bit source? I mean "push" dynamic range in post, I mean get juice from the clipped highlight and the black darkness

    Please understand, I can't fix problems with school math. And it is basic simple school math. Nothing to do with highlights and darkness :-)

    Please just read link with buckets analogy, it has ALL you need (except working frontal core).

    8bit 4k downscaled to 1080 yields a 444 image with "virtual" (for lack of a better term) 10 bit? This virtual 10 bit is good but can be expected to be noisier than native 10 bit?

    Nothing is virtual, it is all real.

  • Wow @Ze_Cahue You are certainly mixing apples with oranges. One has nothing exclusively to do with the other. Two totally separate topics there. @tosvus said it correctly. You could have a very high DR camera that only shoots 8 bit and a very low DR camera that shoots...heck....14 bit. Yes..HOW you shoot your scene helps big time... But whether 10 bit or 8 bit...you can't produce information in the highs or lows that was not there to begin with. That's not what people here are talking about. They are talking about the results from downscaling a HUGE 4K image into a 1920x1080 Full HD image.

  • You could have a very high DR camera that only shoots 8 bit and a very low DR camera that shoots...heck....14 bit.

    It all depends :-) If you are talking about raw for high DR you need good sensor, good ADC (say 12bit or 14bit low noise one) and also low noise. After you go from linear (almost) raw to the non linear space you can use less bits (by using human vision properties). Hence, yep, you can compress original high DR source into 8bit. Photo guys make even extreme HDR that look very weird.

  • OK I'm a confused person, and my computer is crazy when I grade material from the old and good ex1 and get much more DR to push in post if recorded externaly 10bit instead of internaly 8bit. I think its pure Psycho from my mind : )

  • @Ze_Cahue, I am guessing you don't record in 4K 8-bit and downscale to 1080p with your current equipment/workflow?

    You will most certainly see problems when working with 8-bit, and keep in mind that you are likely also working with either uncompressed or much less compressed material from your external recorder than the internal codec.

  • OK. Keep dreaming you can retrieve dead DR from any resolution (4k 5k 6k...) at 8bit downscaled to 2k at 10bit. Lol... Maybe you math guys never worked before with 10bit video. You can retrieve black'n whites pushing the latitude in post.

  • @Ze_Cahue Your mindset is wrong and you are mixing together many variables. It's like saying a rock is heavy, and a car is heavy, so a car is the same as a rock. It makes no sense.

  • @tosvus 10bit isnt only color, it holds more DR too. The theory you guys are talking is right, but only to get smoother color gradation. 10bit is much more than that. Once encoded into 8bit all the whites and blacks become flat white and flat black. You can have a 10k resolution file in 8bit, the whites and black are already flat and all the possible tonality behind them are already washed out, no matter downscale to 10 or 14bit. I'm not talking about color here, I'm talking about the full power of true 10bit. And 10bit holds much more DR than 8bit. Color you can smooth by mixing adjacent pixels of a larger resolution to improove the gradation when downscaled into a 10bit file. But DR once become flat cannot be retrieved.

  • @Ze_Cahue. Of course, but my point is, if the processing is done correctly on the GH4, it will be possible to pull true 1080p 10-bit from a 4K 8-bit source-file. The reason is that the GH4 does the processing in 10-bit, and down-conversion happens last. If I worked for Panasonic, I could write code in 20 minutes that upon the downconverting, would make sure that for each pixel that will be present in the 1080p picture, I take the full 10-bit color information and split it out on 4 different 4K pixels, and make sure when you add them up, they match. This is VERY simple, and will work 100%. No need to worry about anything - you get TRUE 10-bit in a 1080picture (provided you also have a similarly smart resolution down-conversion process).

    Whether Panasonic does this or not is another matter, but it can EASILY be done.

  • @tosvus - If someone writes a transcoding utility that has the conversion accuracy you describe, along with the option of producing either dnxhd or pro-res at the end of it, you would have some very happy campers

  • Question: aren't they getting to the point where Panny's prosumer cams like GH4 threaten to undermine their broadcast cameras? I mean what will the argument be? Buy the broadcast cams because they have a Tally light?

  • @tosvus but to avoid loses, the GH4 must have a nice picture profile, with maximum expanded DR. I mean going to the limits. Because the camera processing could retrieve all the DR while still RAW, before going into a flat 8bit file. After become 8bit, there is not too much to do. See, for example: a highlight in the sky. Even if the camera get a nice code to pickup these 4 8bit pixels into one 10bit pixel, these 4 pixels already would be white, 4 white 8bit pixels going into onde 10bit pixel. So what the meaning of more pixels if these pixels would be all cliped highlights? Native 10bit has more room before these white become a dead flat white.

  • Ze-cahue, you still don't get it and I cannot explain it in simpler terms, so believe what you want.

  • OK @tosvus, I'm not a programmer, but I know how a 10 bit footage behave in post, I work with that from several years. Believe me, your math isnt wrong, but lack of more parameters. I recommend reading other forums that I'm not allowed to link here, I may not use the right words to say how it works properly. Anyway, lets wait for the GH4 and hope for a good processing.

  • @Ze_Cahue I am not a hard core programmer either, but I do write scripts and build complex 3d world models as well as work in audio. What tosvus is explaining is simple to understand and you appear to be assuming otherwise, I can also tell from your posts that English is your second language and maybe that is the problem. What he is saying is that the GH4 does the initial processing in 10 bit, therefore the intital 10 bit data is there in he first place in the GH4. So the gradation in highlights etc...will be there in 10 bit in the first place, and it downsamples to 8 bit...not the other way around...( 8 bit in the Camera trying to be 10 bit) and thats what you appear to be assuming, and others appear to assume this also. When you say "missing data" " all white pixels"etc...you are thinking it's initially baked in 8 bit in the GH4. He's saying its 10 bit FIRST...and it has to be (with or without the brick) so the DATA is there, its a simple matter of conversion etc..not hard to understand. Cheers

  • Lots of GH4 samples over there and nobody is getting 10bit color from a 4k 8bit file. Maybe you guys should teach them how to do it properly. I also did my try and got tipical 8bit banding while trying to grade a "magical 1080 444 10 bit" from a 4k 420 8bit.