G
Guest
Guest
The other thing about all these review and first look videos, shooters are getting so good, they can make any camera shine. The talent pool has increased significantly. It is quite awesome.
Sony A7-II: XAVC-S, slog-2, and now... IBIS
- Thread starter Samuel H
- Start date
Grug
Veteran
Can you adjust picture profiles with the various A7 models? SLOG2 on 8-bit just doesn't seem like it'd offer enough tonality in the midtones for a good image. I feel like you'd really want to add a bit more contrast and saturation into the SLOG2 image to yield something reasonable after grading.
Ah, but you can - at least in theory. The way forward would be for a sensor to have EXACTLY 4x the dimensions of the 1080 frame on each axis.
In other words, 4x1920 horizontally, and 4x1080 vertically. Do the maths, and you get 33,177,600 photosites. Allow for photosites around the edges and you're close to 36Mpix. I'd also expect a still/video sensor to have an aspect ratio of 3:2 (for stills) which gets cropped to 16:9 for video, which would mean a corresponding increase over the 33.2Mp to allow for that.
Reasoning is that you read in blocks of 4x4 - which gives 1920x1080 blocks. And with 4 red, 4 blue, and 8 green photosites per block, the reading process simply bins all the charges of corresponding colours together to give three values per block for R,G,B. So you get a 1080 frame directly with no need for "downconversion" as generally understood. It's comparable to the way the C300 gets it's output, but in that case we're talking about 2x2 blocks, here it's 4x4 - but the same principle. And with such a chip, do exactly as the C300 does (use 2x2 blocks) and you get 4k directly.
As far as lowlight performance goes, then the binning process should mean it's at no disadvantage here. Binning the electric charge from 4 small photosites should give the same results as from a single photosite of 4x the area.
Samuel H
Major Contributor
4x1920 and then 3:2 aspect ratio is 7680x5120, or 39 Mpix. Downsampling may be simple, but converting the analog readings into digital data fast enough, and getting all that data out of the sensor fast enough, definitely is not. You'll need line-skipping, and probably a lot of it.
Look at the a7s and its awful rolling shutter. Full-readout at 4K on such a big sensor is already testing the limits of current tech.
The only way to get that is with in-sensor analog binning. Some sensors can do that. That would be lovely...
* You wouldn't need to do a complex debayer (as you point out, you can go for the C300 method).
* "All you need to add to the sensor" are vias and transistors to pool together the charge from the same-color photosites of each block when shooting video instead of stills, then everything else is relatively standard (and you saved 1/4 of the signals, so you definitely can do 4K-out-of-8K, with the same amount of ADCs and LVDS lanes as the a7S has, and have a 5D3-like rolling shutter)
The problem is: much easier said than done, I'm afraid...
Look at the a7s and its awful rolling shutter. Full-readout at 4K on such a big sensor is already testing the limits of current tech.
The only way to get that is with in-sensor analog binning. Some sensors can do that. That would be lovely...
* You wouldn't need to do a complex debayer (as you point out, you can go for the C300 method).
* "All you need to add to the sensor" are vias and transistors to pool together the charge from the same-color photosites of each block when shooting video instead of stills, then everything else is relatively standard (and you saved 1/4 of the signals, so you definitely can do 4K-out-of-8K, with the same amount of ADCs and LVDS lanes as the a7S has, and have a 5D3-like rolling shutter)
The problem is: much easier said than done, I'm afraid...
Last edited:
Which is what I was thinking of. The intent was to make a general point ("Ah, but you can - at least in theory. The way forward would be.....") to refute the principle of what was previously said ("You can't have a 36mp sensor that shoots great 1080p video ......").
My apologies if I inferred I was thinking of a "here and now" sensor. The points were intended to refer to a 33Mp sensor in principle (or 39Mp as you say to cater for 3:2 stills.)
Ebrahim Saadawi
Carbonite Member
It's just a very high resolution 36mp sensor on full frame will not have as good as a 12 full frame lowlight performance, and also it will not have to downsample as much therefore sharper image. Not to imply that a 36mp sensor can't shoot great 1080p, I am testing the D810 and it's blowing my away, but not nearly as good as the A7s neither is resolution or lowlight performance.
This will always be a compromise we'll have to make. High resolution stills or better video quality and lowlight performance.
This will always be a compromise we'll have to make. High resolution stills or better video quality and lowlight performance.
Samuel H
Major Contributor
Yes, I got your point. I was just underlining that it's not just adding more megapixels and some extra processing, you actually need a new kind of functionality, and a physical redesign of the sensor at a very low level.
BTW this kind of design would make the tradeoffs Ebrahim mentioned a lot less painful: you would get nearly the same low light performance, for example (comparing 4K-from-4K with 4K-from-8K-with-analog-in-sensor-binning) (you lose some photons because of the smaller microlenses, but that's going to be one stop at most, and mostly in the corners, i.e. more vignetting wide open, as you can already see comparing a7r vs a7s).
But it's not easy...
Last edited:
Celluloid Dream
Active member
I think the ideal 3:2 high resolution still camera sensor for 4K video would be exactly 44.7 megapixels or 8192x5461 pixels.
That is to more easily implement 4K video and with high quality, while still producing outstanding high resolution still images
Why 8192x5461 you ask? Because the 4k professional production standard is 4096x2160. If you have an image sensor of 8192x5461 it is a much simpler to bin two pixels to one and end up with a video image size 4096x2731, which the camera will crop away to get exactly 4096x2160...
That is to more easily implement 4K video and with high quality, while still producing outstanding high resolution still images
Why 8192x5461 you ask? Because the 4k professional production standard is 4096x2160. If you have an image sensor of 8192x5461 it is a much simpler to bin two pixels to one and end up with a video image size 4096x2731, which the camera will crop away to get exactly 4096x2160...
4x1920 and then 3:2 aspect ratio is 7680x5120, or 39 Mpix. Downsampling may be simple, but converting the analog readings into digital data fast enough, and getting all that data out of the sensor fast enough, definitely is not. You'll need line-skipping, and probably a lot of it.
Look at the a7s and its awful rolling shutter. Full-readout at 4K on such a big sensor is already testing the limits of current tech.
The only way to get that is with in-sensor analog binning. Some sensors can do that. That would be lovely...
* You wouldn't need to do a complex debayer (as you point out, you can go for the C300 method).
* "All you need to add to the sensor" are vias and transistors to pool together the charge from the same-color photosites of each block when shooting video instead of stills, then everything else is relatively standard (and you saved 1/4 of the signals, so you definitely can do 4K-out-of-8K, with the same amount of ADCs and LVDS lanes as the a7S has, and have a 5D3-like rolling shutter)
The problem is: much easier said than done, I'm afraid...
A counterpoint - Samsung NX1 has 28 MPX on an APC sensor, which projects to ... ~ 65 MPX (?) on a full frame camera. At that point, one should be able to lower the pixel count while increasing the pixel size to accommodate an 8K/8K+ type of a sensor with 40-50 MPX.
PS. Samuel, you must be ecstatic with more Sony Alpha series lens announcements.
PS. Samuel, you must be ecstatic with more Sony Alpha series lens announcements.
joe1946
Veteran
I use the 36mp A7r mostly for large prints up to four feet.
https://farm6.staticflickr.com/5598/15360260540_65e2eb0d11_o.jpg
G
Guest
Guest
Holy megapixels batman!
G
Guest
Guest
I am a bit divided. Having a lowlight camera like the A7s is a great option to have, but 3200 native iso, what does that do when you want to shoot during the day? That sounds like a lot of ND, or some funky s-log at reduce iso's.
So, I may end up liking the A7II with s-log, for a daily pocketable camera, which I am in the market for. We'll see, I guess. Any footage videos yet? At this point all of the A7 footage looks less exciting.
So, I may end up liking the A7II with s-log, for a daily pocketable camera, which I am in the market for. We'll see, I guess. Any footage videos yet? At this point all of the A7 footage looks less exciting.
Last edited:
dustylense
Veteran
No, that doesn't necessarily follow - as I said earlier: "Binning the electric charge from 4 small photosites should give the same results as from a single photosite of 4x the area. , and as Samuel H puts it:
Samuel H said:
The whole point about what Samuel H and I are talking about is that it doesn't "downsample" at all - not in the generally accepted way anyway. If the sensor is an exact multiple of the frame dimensions (and ideally an even one) binning becomes a very valid technique. Each binned block directly corresponds to a single pixel of the output frame, so no need to do a conventional deBayer or scaling. That gives full video resolution with low power consumption because of the minimised processing.
The proof is in the C300, but because there it's only twice the frame dimensions in each axis it means "only" about 8.3Mpx. So great for 1080 video, rather low for stills. Next even multiple after 2x is 4x - which would indeed be capable of good video performance, but also high quality stills as well.
Yes to the comments from Celluloid Dream about true 4k (as opposed to quad-HD) needing 8192x5461, so yes - 44.7Mpx would be needed for that. But maybe that's looking at running before walking.....? For now I'd be happy to see such a 39Mpx 3:2 sensor with binning capability.
G
Guest
Guest
We started talking about this over hear: http://www.dvxuser.com/V6/showthread.php?324931-8K-DSLRs
Basically, seems feasible. And I really only care to have standard size motion picture 4k, so everything else could be approximations. even still, they match up pretty well with format sizes and native resolutions:
resolution based on ~1.85-1.9:1 format with (standardized format size for reference)
-SD ~ 1/3" (4:3) 4.8 x 3.6
-720p ~ 1/2" (4:3) 6.4 x .4.8mm
-HD ~ 1"-2/3" (16:9) 9.587 x 5.83mm
-2K = 12.445 x 6 mm (s16 = 12.5 x 7.41 mm)
-3K= 17.3x13.3mm = M43
-4K = 24.89 X 12 mm (s35 = 24.89 x 12 mm)
-4.5-5K = APS-H equivalent
-6K = 37.335 x 18 mm (ff35 = 36 x 24 mm)
-8K = 49.78 x 24 mm (65mm = 52.51 x 23.07 mm)
All from one sensor!
In theory, the real beauty of the sensors being thought about in this thread are the SIMPLE way they give high quality video, yet have a sensor with enough photosites to satisfy demanding stills work.
That's down to them having that direct mathematical relationship to final video frame size. (Number of photosites is an even multiple of video frame size horizontally and vertically.)
The moment you start going into any crop mode (other than 16:9 video from a 3:2 sensor for stills) then you lose that simplicity and the camera needs to use far more complex processing. (Full deBayering and scaling.)
Far more satisfactory and simpler to use the whole sensor all the time - and use 2x2 binning to get 4k, 4x4 binning to get 1080.
The exception to that may be a 2x crop, to use a quarter of the sensor, when full 1080 could be easily derived by the 2x2 method from 3840x2160.
The physical size of the sensor is immaterial as long as the photosite count is one of the "golden" numbers (such as 39Mpx) though obviously the smaller the overall size, the smaller the photosites.
That's down to them having that direct mathematical relationship to final video frame size. (Number of photosites is an even multiple of video frame size horizontally and vertically.)
The moment you start going into any crop mode (other than 16:9 video from a 3:2 sensor for stills) then you lose that simplicity and the camera needs to use far more complex processing. (Full deBayering and scaling.)
Far more satisfactory and simpler to use the whole sensor all the time - and use 2x2 binning to get 4k, 4x4 binning to get 1080.
The exception to that may be a 2x crop, to use a quarter of the sensor, when full 1080 could be easily derived by the 2x2 method from 3840x2160.
The physical size of the sensor is immaterial as long as the photosite count is one of the "golden" numbers (such as 39Mpx) though obviously the smaller the overall size, the smaller the photosites.
Samuel H
Major Contributor
Wow that's really detailed
G
Guest
Guest
The idea that me and a few others were talking about, is to see if theoretically cropping matched round resolutions to standard formats. The simple answer is no, but the interesting thing is how close it actually does match up. Being able to switch perspectives by cropping would be a game changer.
joe 1008
Veteran
Even if the A7II would be two stops slower than the A7s the internal stabilizer could compensate in run and gun situations: The fastest lens for the A7s with IS (using an adapter) has an aperture of 2,8 - with the A7II you can shoot handheld with a ton of f1,4 lenses or even faster glass.
G
Guest
Guest
Sold!