Bassman2003
Veteran
We will have to wait until the GH4 is out in the wild. Until then it looks like shooting in 4k for everything outside of 60p/50p is the way to optimal quality.
Panasonic 4k camera system info & updates
- Thread starter PappasArts
- Start date
John Caballero
Veteran
It gets really boring at this point. We are officially at the point where all the "experts" start "contributing" their conjectures for a bunch of useless non-conclusions. It is amazing how the timeline of events and blabber repeats itself time after time. Wait until the camera is in the hands of users for a proper analysis of how it performs. Misinformation is an enemy.
joe 1008
Veteran
A few pages back a review stated that they were only getting about 30 minutes of 4k 100mb/s footage.
I thought we were looking at 100 megabits per second with 4k, which would translate to this for a 64gb card: 100/8 = 12.5mb per second. 12.5 x 60 = 750mb per minute. 64,000mb/750mb = 85.33 minutes of video. Quite a bit different from 30
What am I missing? (or is the review off)
I thought we were looking at 100 megabits per second with 4k, which would translate to this for a 64gb card: 100/8 = 12.5mb per second. 12.5 x 60 = 750mb per minute. 64,000mb/750mb = 85.33 minutes of video. Quite a bit different from 30
What am I missing? (or is the review off)
John Caballero
Veteran
So is my impatience.
LOL! I know the feeling!
legrevedotcom
Veteran
Wtf at the price…. half price of a new 5D3.
Not quite. Binning is only 'lower quality' in the sense that you lose a bit of fine detail because the photosites are in essence, bigger. But binning has one enormous advantage - the signal to noise ratio is four times better, because you're combining four photosites into one. Low light performance is dramatically improved. You can clearly see this with the GH2 and 3. Notice how much noisier ETC mode is? That's the native, 1:1 non-binned picture.
Last edited:
No - whole sensor or centre crop MUST make a huge difference in performance. That's down to geometry. What is being described as "2x2 binning" (forming an R,G,B output pixel directly from a 2x2 Bayer block, red directly from red, etc) will give a very high quality HD output - but ONLY from a 3840x2160 photosite array. Understanding that is crucial. It's because you need 1920x1080 blocks - so 3840x2160 must follow.
Try to do it with the higher count of the full sensor, and how do you bin the photosites together? (Remember "binning" in it's true sense only works amongst photosites of the same colour.) From the specs in stills mode, it seems the 16:9 dimensions of the full sensor are 4608x2592. Perform "2x2 binning" on that, and you'll get a full colour image of 2304x1296. Well OK - but to then get the desired 1920x1080 that would have to be downconverted, and doing that well, in real time, is difficult, and beyond the abilities of nearly all DSLRs so far. Consequently they employ a means of skipping or combining photosites, typically binning to a 4x4 block, which gives an intermediate image of 1152x648 - and upconverting is a far easier process than downconversion.
Binning may be seen as "a lower quality way of down converting the sensor" only when compared to a "proper deBayer". If you want 1080 off a 4K sensor, it'll give you very good quality 1080 - but you need a proper deBayer to get something worthy of being called 4K. Getting quality 1080 off a sensor with a photosite count that's not a mathematic multiple of 1080 is what causes problems. That's why it's wrong to say "binning is binning".
Thomas Smet
Veteran
Yes but part of that is down converting in general. Down converting a 4k image in post to 1080p also removes a lot of noise. Any time you take a 1:1 pixel readout it is going to have more noise than doing some form of down conversion either via binning scaling or whatever. Scaling down with a good down scaling formula can also greatly reduce noise by merging pixels. Same reason why down scaling can give you 10bit and 4:4:4. You create a new super pixel that is a combination of four pixels. The problem with binning is it is well known to not be as good as a good quality software down conversion. This is part of the reason why DSLRs tend to have more aliasing and moire and a false sense of detail compared to a traditional video camera. Shooting 4k and down converting will yield much better down conversion results in terms of clean detail with fewer image artifacts.
You can get this effect right now with the GH3. Take a noisy HD image and down convert to 960x540. You will get close to 10bit color, 4:4:4, and much cleaner overall quality with less noise. You can see the advantage of a 4k down conversion by taking a GH3 still image and doing the same thing down to 1080p. Did a test the other night by taking a GH3 still image cropping to 4096x2160 rendering to a H264 video at 100mbps 4:2:0 and then scaling that rendered video down to 2048x1080 in After Effects which will process the video and give you 4:4:4 10bit with less visible noise.
Thomas Smet
Veteran
Yes there will likely be a difference but I doubt by as much as you think. You may not notice it as much. Heck even down converting 4k may not yield results that everybody will notice or appreciate. We are talking about splitting hairs here and something 99% of the people in the world would never notice anyway. This is what I mean by debatable. I used to write image processing software and while there can be a difference it is likely not going to make a major difference to most people. The only way I honestly believe to get a cleaner image outside of binning is to sample a much larger readout and do the scaling yourself. We used to do this with RED ONE footage as well. Always said the best HD I ever shot was with a 4k camera.
No - but well, there's "binning" and there's binning......
Done on the 2x2 basis (for which I prefer the term "direct read", as only two green photosites really get binned) it's very good - and the C300 proves that.
But it's a special case (starting with 3840x2160), so the process directly yields true 1920x1080 and no need for any further scaling. That's good. You CAN get just as good from any still camera sensor (or the GH4 here) by cropping to 3840x2160.
No again. The problem for DSLRs is that there are too many photosites to do the 2x2 process as above. If you try you don't end up with the 1920x1080 image - you'd have to put in a downconversion stage.
In theory - yes. But you are presumably talking about using a computer to do the downscaling, and not necessarily in real time? In other words, huge processing power, no power supply limitations, and no time limitations to process each frame.
Move to a DSLR and it's a different world. Limited electrical power, limited power dissipation capacity, size, cost, etc all have to be kept down - and each frame has to be processed in real time! That just has not been possible in real world DSLR cameras, at expected price points, with current technology. It's not impossible to design such into a camera - but the point has been that the cost would have been so high (both money, and size, power consumption etc) that it would have been a compromise to it's primary function - taking still images.
Again, I won't dispute what you say - but again, you're presumably doing it with a computer and with no constraints of power or even that it has to done at real time framerates?
Taking the approach you suggest, and applying it within a DSLR, would add a lot to the cost etc. It is broadly how such as the F3, Alexa etc process the sensor - but look at the difference in cost, power consumption etc between such and a DSLR in video mode!?!
It becomes difficult to generalise, but arguably the best compromise still cameras use for video, is to do something like a 4x4 bin and UPCONVERT as said above. A far easier process than downconversion.
But..... it means the resolution must then be limited to a quarter of the sensor resolution on each axis, not a half - hence the typical 1152x648 I quoted above.
But use a 2x2 "bin" and 3840x2160 and you'll get full 1920x1080 with simple (capable of being used in a DSLR) processing. That's why I've previously described 3840x2160 as a "magic" number for sensor resolution in this context. Whether achieved by centre cropping or native sensor resolution.
Thomas Smet
Veteran
Of course I'm talking about doing so on a computer. The discussion isn't about the best method for a camera to process in RT but how shooting 4k can yield superior quality.
I am well aware a camera can only do so much in RT. This is why some of us like the option of being able to shoot 4k. It allows us to determine how we want to process our material. We have two great options with the GH4. Handle our own processing in post for superior quality or let the camera handle it.
Without any solid proof I still stand by we are splitting hairs here. I doubt there would be an earth shattering difference. At least one that few would recognize the difference. Again I am not arguing that there wouldn't be a difference just that it may not be a huge amount. Of course we will never know since the GH4 will not be able to do this.
John Godden
Well-known member
FYI: Averaging 4 'bins' will increase S/N (at best) by a factor of 2.
joe 1008
Veteran
Here a slightly different approach to the downscaling-discussion: As a lot of GH4 footage will be shown on the web, one could leave all the downscaling to the provider (vimeo, YouTube etc.) or to the final consumer. In my experience such 4K footage watched on my HD screen looks much better than any previously downscaled and recompressed footage. I see a real pixel por pixel image, all compression artifacts are gone. Just watch the GH4 Yucatan-footage http://www.youtube.com/watch?v=hHKJ5eE7I1k reproduced @ 4K on a HD screen and you'll see how good it works. Also impressive in that way are some of the clips in Philip Bloom's postcard collection on vimeo shot on his 1Dc http://vimeo.com/82756917. In that case vimeo did the downscaling and it did so impressively good.
True North
Well-known member
I just watched the Panny video on my ipad and you are right, it looks absolutely stunning
I think you bring up an interesting point and perspective, whether we like it or not, the vast majority of anything posted or shot in 4K or UHD is going to end up being down scaled and decompressed anyways but any number of devices or software. 99.9% of consumers and even most people on this board don't have a medium to transmit or reproduce true 4k from start to finish.
Michael Schrengohst
Well-known member
Not yet - I struggled for years to get clients to use DVD…..then I struggled for years to get them off DVD and playback using laptops and files…..now I have some clients asking
why the videos are only 720p??? Heh, some clients say the 1080p videos I do is overkill and some complain that they are not sharp enough…..We will always have these struggles.
But 4K is about to go mainstream because no one will go back to buying 1080 camcorders and DSLR's. And in a few years it will be hard to find 1080 displays. I can't wait
until a 60" 1080p LCD can be had for $89.
OK, I think we're talking cross purposes - two cross discussions in the same thread.
My original post (in response to Samuel) was regarding the difference between real time 1080 modes for either full sensor read, or derived from a 3840x2160 crop. In that case, the cropping mode will almost certainly give a far better result than full sensor. And no, that difference won't be splitting hairs.
But talking about the 4K mode, then yes, I'm in full agreement that is a great option - even if the end product won't be 4K. I don't see a huge difference between a software downconversion of that and in-camera binning of the 3840x2160 crop - but I do expect to see both of them substantially better than the in-camera full sensor 1080 mode.
As far as the "if it's going to be on the web....." comments, then maybe aliasing is still going to be highly relevant here. The whole point about why aliasing is bad is that it may not be very visible on "real" pictures, but can have a huge effect on high compression codecs. You can have two original sequences which may look similar in quality, but one may look substantially worse than the other when compressed to the same amount. Which is exactly why broadcasters may seem to be over fussy about not approving some cameras, in spite of their first generation output seeming OK.......
(If anyone wants the brief (simplified) reasoning, it's because if an object shows aliases, and moves within the frame, then the aliasing travels in the opposite direction! That confuses algorithms in compression software - and wastes bitrate. The effect is very obvious if you've ever seen a resolution chart block which is aliasing, and panned the camera from side to side. Which is the simple way to tell on a chart what resolution is real, what is aliased. This helps to explain why the whole topic of aliasing is so much more important in video images than still photographs.))
joe 1008
Veteran
Interesting point.
I have a retina iPad and it won't let me do more than 720p. How are you watching this?