It shoots stills at only 3fps, so readout is about as slow as slow can be. There's no way they can get a good video image out of that sensor.
new Pentax 645D medium format with 50mp Sony CMOS sensor with video recording mode ?
- Thread starter joe1946
- Start date
Kolor-Pikker
Well-known member
The sensor itself is only part of the story, the actual implementation by each manufacturer can vary widely, and besides, 3fps is the still-shooting speed right? There aren't many processors that can maintain cost-effectiveness and quality output when dealing with such massive resolutions.
When Phase One designed their implementation of the Sony sensor, I highly doubt they tuned any aspect of it that would be suitable for video recording, but other companies might focus their R&D here as well.
Jorn van Oostende
Active member
I remember when Phase One and Imacon backs would not shoot faster then 1shot per 1.5sec. And lets not forget the Sinar scan backs that needed sometimes an hour per shot, lots of down time during those shoots...
Samuel H
Major Contributor
thedarkhorse
Member
now when people are discussing the cons of new cameras they won't be complaining "it's not full frame" they will have to say "it's not medium format". Throw away your 5ds, they are utter garbage now.
Kolor-Pikker
Well-known member
Well, what all medium format film had in common was a 6cm width, with the length used per shot being determined by the camera you loaded it into; so what you really need to do is reach 54x??mm to be considered "medium format".
Not necessarily. Sam H has already sort of answered it, but if you want to produce high quality video from a high pixel count sensor, the trick is to have one of the "magic" values of pixel count. Which tend by and large to be exact multiples of the required final resolution - which avoids up or downscaling, and can avoid intensive deBayering.
Most obvious is to use 4096(3840)x2160 - and form each output pixel directly from a 2x2 Bayer block. But that is great for video - but maybe too low a resolution for high quality stills. I'd say you can get to the next "magic" number by doubling everything - so basing it on 8192x4320 (about 35 Mpixel). For 1080, you now form an output pixel out of a 4x4 Bayer block instead of 2x2. Since the sensor is 4:3 aspect ratio for stills use, you are therefore looking at a total count for the sensor of about 8192x6144 - or about 50 Mpixel. Hmm - funny that......
Look at the specs, and it lists the actual max stills resolution as 8256 x 6192 - so you get the "magic" horizontal count with a crop of only 1.008 relative to the full sensor! I don't believe that's an accident.
It's exactly the photosite count that I'd expect to see to achieve the Holy Grail of a high res sensor for stills, and at the same time to give uncompromised video performance with fairly simple processing.
That's the theory of why a 50Mpixel sensor SHOULD be good. What is actually happening here, I don't know. What's needed is for all the photosites in each 4x4 block to be read out, and corresponding R,G,B photosite values to be binned to form 3 values for R,G, and B. That shouldn't take any longer than the same process from a 4096x2160 sensor - that's the beauty of binning. And because each output pixel is being formed by such an averaging process, there should be very little aliasing - the averaging is effectively doing a low pass filtration job.
But really, I'd expect this camera to be bought primarily by people looking at it first and foremost for stills use? It should produce excellent 1080 video - but if that's all you want, there are probably better buys around.
joe1946
Veteran
The new Pentax 645Z is a fraction of the cost of the Phase One and Hasselblad digital backs.
http://www.ricoh-imaging.co.jp/english/products/645z/feature/04.html
http://www.bhphotovideo.com/c/product/1045517-REG/pentax_16599_645z_digital_slr_camera.html
http://www.bhphotovideo.com/c/product/1034017-REG/hasselblad_h_3013668_h5d_50c_dslr_camera.html
http://www.dpreview.com/news/2014/01/24/phase-one-announces-iq250-50mp-cmos-medium-format-back
dustylense
Veteran
Well, whoever buys this for video before stills is a ding-dong. As a stills camera this device should be welcomed by stills shooters who want larger than 35mm looks and 50 megapixels of imaging. All for less than 10K? Come on, thats a deal provided the images hold up. And maybe the video in static shots will work out nice. But this is a stills camera first with a video function added.
Samuel H
Major Contributor
I hadn't noticed, but funny coincidence indeed!
But I wouldn't share your optimism: since they need to accelerate the read-out time a lot to get video frame rates, my guess is that it will read one of every two or four rows of pixels, and average every 4x1 or 4x2 values to get each final 1080p pixel. If it is 4x1, it will have pretty bad aliasing.
I assume you have sarcasm font on as well, when you write "coincidence...."
Not necessarily, that's the beauty of binning. (A term which often gets confused and wrongly used for skipping, which is what you refer to.)
In a 4x4 Bayer block, you'll have 8 green photosites, 4 red, and 4 blue. When the block gets read, then as I understand it, the charges in the 8 sites corresponding to green all get read at the same time and accumulated together into the same "bin" - hence the name. Same for red and blue, and the three values then correspond to the R,G,B of the output pixel. Point is that reading 8 sites (accumulating together the electrical charges) into the same bin should be far quicker than reading 8 sites into 8 bins and dealing with them. It could be comparable to reading 2 sites into a bin in the 2x2 arrangement.
To put it another way, it may be the number of bins that mostly determines speed - not so much number of photosites.
In such a 4x4 bin, then aliasing ceases to be a problem in the way it would be with the skipping arrangement you suggest. Any detail finer than the system resolution will be fine compared to block size, hence the averaging. I'd suggest that if it doesn't bin like that, but skips more in the way you suggest, it's more likely to read two lines, skip two, then ignore two, read two across each line. The result is that you'd get a 2x2 block read out of every 4x4 block - the other 12 photosites being ignored.
Real point is that a 50Mpixel sensor *COULD* be very good indeed for video use as well as high res stills. Obviously, it needs to be relatively large to keep a decent photosite size (which here it is), but I think the same basic principle could still be OK with FF 35mm, maybe even s35.
Samuel H
Major Contributor
Of course, if Sony has mastered in-sensor binning with full-readout, in a way that allows them to greatly reduce sensor readout times(*), then this will deliver incredibly sharp video with no aliasing/moire.
(*) In theory the best you can do is a linear reduction, which would mean that if the sensor can spit out 50 Mpix at 3 fps, then it can spit out 2 Mpix at 75 fps. But that's actually never the case, because there's an overhead that's not reduced by the binning, and because the binning usually adds extra steps that take some time.
The only indication I have that Sony may be doing this is that the a7s seems to have pretty bad RS in some videos (the infamous RS test from NAB) but not nearly as much in others (Scotland, which was shot at 1080p), and this could mean that the RS is much worse at 4K than at 1080p, even if both use information from all the photosites. Or maybe Scotland just chose the clips without obvious RS issues.
We'll know soon enough...
Off the top of my head, is it possible that in stills mode the limiting speed factor may not be sensor read, but debayering, writing to card etc etc? No idea if that is the case, but extrapolating just from still burst speed may turn out not to be an accurate way to predict video characteristics?