[falconeye]

that's precisely what the canon C300 does: downsample 8mpix bayer to super sharp 2mpix 4:2:2 video
[...]
but keep in mind that storing those 8mpix in RAW, as you suggest, can be problematic: those 360 Mbyte/s

Thanks for the reference. Yeah, but it isn't the price point of the VG series. The point I was making is that nothing prevents Sony from doing this in their VG series. The 8MP downsampling isn't that expensive ...

btw, I did not suggest to record 8MP RAW movies. Personally, I would output a stream of weakly compressed 12Bit 2MP JPG XR images. Or whatever the movie industry has as equivalent.

The 360 MB/s stream is meant to exist just between the sensor and DSP.

[Samuel H]

oh, right, then: that's indeed very very feasible
the C300 does exactly that with its 8mpix sensor and a Digic DV III processor that can also be found on $200 consumer camcorders, so, if you can get the data out of the sensor fast enough, everything else is definitely not a big feat

regarding codec, both Canon and Sony have nice mpeg-2 8-bit 4:2:2 codecs working at 50 Mbps (Canon-XF and XDCAM-HD422), I guess going 12-bit would require 75Mbps, which sounds very reasonable

if you want a readily available sequence-of-stills format, it can be DCP (digital cinema package): that's what digital projectors at theaters use most often, and even though it's thought out as a delivery format, I think it's good enough for acquisition too (it's a string of JPEG2000 files with up to 12 bits per channel) - it may be very messy to work with in post, though
Digital Cinema Package - Wikipedia, the free encyclopedia

for the K-01, I think all we could ask for is H.264 or mpeg-2 with 8-bit 4:2:2 at a high bitrate, if a hack can achieve that (the GH2 H.264 hack gets a very high bitrate, but it's still 8-bit 4:2:0)

[falconeye]

regarding codec, both Canon and Sony have nice mpeg-2 8-bit 4:2:2 codecs working at 50 Mbps (Canon-XF and XDCAM-HD422), I guess going 12-bit would require 75Mbps, which sounds very reasonable

I think to go beyond 8 Bits is very important with supersampling 8MP APSC or FF cameras. E.g., the K-5 sensor has 14 EV dynamic range at 8 MP (DxO definition) and at supersampled 2MP this increases to even 15 EV. So, an 8 Bit output would loose about half the signal at low ISO from the current high DR sensors Sony makes.

It isn't that big of a loss with subsampled video as in current HDSLRs. But with a more dedicated sensor, it will and MP4 or MJPG won't be appropriate anymore for post processing lattitude. Of course, some of the DR could be captured by having a very low contrast profile.

[Anvh]

Cinema DNG maybe?
I've no idea how heavy that is on the processor.

Now that i think about it if the downsampling isn't processor extensive the K-01 might be able to pull it off, 1080p24/25 in cinema dng.
Not sure if 30 can be reach or higher...

---

Last edited by Anvh; 02-14-2012 at 07:08 PM.

[falconeye]

Now that i think about it if the downsampling isn't processor extensive the K-01 might be able to pull it off,

I thought I mentioned it. The sensor readout bandwidth is the bottleneck. No way the K-01 can do it. Faster sensor readouts produce heat and are expensive due to their high pin count. The best solution would be a binning function embedded into the A/D column parallel converters which are part of the newer Sony sensors.

[fuent104]

Cinema DNG maybe?
I've no idea how heavy that is on the processor.

Now that i think about it if the downsampling isn't processor extensive the K-01 might be able to pull it off, 1080p24/25 in cinema dng.
Not sure if 30 can be reach or higher...

High-end cinema cameras record one raw image file per frame. the RED Epic, in HDRx mode, actually shoots three per frame, at different iso settings, to be combined for maximum dynamic range. The file formats are different for different camera systems. RED uses Redraw, which I believe is saved as .r3d files. The Arri Alexa can record raw files, but only if you record to an external recorder. It is considered by some to be the best digital camera in the movie business. Without the external recorder, it can only record at 1080.

[asw66]

All this talk about down-sampling and so on has got me wondering if it is feasible to replace both the Bayer filter and AA filter with a dynamic (e.g. LCD based?) filter that can be programmed to handle colour management and moire reduction, as well as selectively blocking out pixels if down-sampling is required.

I have no idea if this is remotely possible, but it could be very cool if it were indeed feasible.

---

Last edited by asw66; 02-15-2012 at 05:53 AM. Reason: jargon fix

[Daemos]

I just want the video of the K-01 to be better than that of the Panasonic GH2 (with hacks). IE. Supersampling down to 1080p, no rolling shutter, no use every 3rd pixel (like the 5D) etc...

I've read some reviews where reviewers rate the 1080P video (with firmware hack) of the GH2 to be superior to that of the 7D and the 5D MKII, and in terms of tech on how they get the video similar to the C300...I don't expect C300 quality, but I want better than GH2 quality

[Samuel H]

in the areas where it shines (resolution), the hacked GH2 is definitely a beast, easily comparable to the C300
I wouldn't expect the K-01 to get such sharp images
there's more to IQ than resolution, though, and at least in stills mode the K-5 has nearly 3 stops higher DR than the GH2...
DxOMark - Compare cameras side by side

but, as I said before, for this to matter you have to check all the minimum requirements first, and maybe the codec will not pass the cut

[liukaitc]

in the areas where it shines (resolution), the hacked GH2 is definitely a beast, easily comparable to the C300
I wouldn't expect the K-01 to get such sharp images
there's more to IQ than resolution, though, and at least in stills mode the K-5 has nearly 3 stops higher DR than the GH2...
DxOMark - Compare cameras side by side

but, as I said before, for this to matter you have to check all the minimum requirements first, and maybe the codec will not pass the cut

again DR in still is different in video
there is no DR advantage of k-5 sensor in video
check out the data from great camera shootout 2011
d7000 (same sesnor as k-5) only got 10.2 stop DR in video

[TaoMaas]

How much some of this stuff matters depends upon a person's end usage. Are we shooting for the family reunion....or to be shown in movie theaters?

[Samuel H]

@TaoMass
movie theaters!! at least hopefully, eventually, some day...
I wouldn't think there will ever be a "cult video camera" for shooting family reunions, as long as you can do it with your mobile phone too

@liukaitc
I know; that's because the canon DSLRs have better tools to decide how what-the-sensor-captures gets baked into the video stream
some variants of my very own Flaat picture styles for Canon DSLRs deliver over one stop more DR than what was recorded in those tests (they do so by pushing up the shadows, and this introduces noise, and reduces the color space available for midtones and highlights, which again means more noise, but hey, it's there if you need it)
Flaat Picture Styles for Canon DSLRs - www.similaar.com
the GH2, like the Nikon, is not very flexible in this respect; I still don't know how the Pentax will fare, but I've asked around in this forum, and while Pentax doesn't seem to have a utility like Canon's Picture Style Editor, it seems to have relatively powerful in-camera controls
https://www.pentaxforums.com/forums/pentax-beginners-corner-q/174744-custom-image.html

[Daemos]

in the areas where it shines (resolution), the hacked GH2 is definitely a beast, easily comparable to the C300
I wouldn't expect the K-01 to get such sharp images
there's more to IQ than resolution, though, and at least in stills mode the K-5 has nearly 3 stops higher DR than the GH2...
DxOMark - Compare cameras side by side

but, as I said before, for this to matter you have to check all the minimum requirements first, and maybe the codec will not pass the cut

I hope the K-01 is good as the cost of good lenses for the GH2 for video shooting is very cost prohibitive...where vintage pentax lenses are abundant and affordable

It's pretty easy to get a nice fast k mount 35mm prime (~50mm eq) where it's expensive to get a nice fast 4/3rds 25mm prime

[falconeye]

again DR in still is different in video
there is no DR advantage of k-5 sensor in video
check out the data from great camera shootout 2011
d7000 (same sesnor as k-5) only got 10.2 stop DR in video

I've no idea what they measured. But chances are high that it is BS.

Dynamic range in JPG (and HDSLR video is just the compressed stream of them) depends on the contrast / shadow preservation settings chosen in the menu and little else.

But this doesn't mean that K-5 is above average in video. Just that the test needs closer inspection in order to be cited. And you don't even provide a link to the source which is really bad practice ...

[Samuel H]

@falconeye: given who ran those tests, given how much time and money they spent doing so, and given that they got an emmy for the 2010 edition of that "documentary", I'd trust that they did the best that could be done, or very close to that

my understanding is that they measured how much of the DR captured by the sensor could actually be saved when recording in video mode

as in jpg shooting, you have to bake in a picture style (custom style in Pentax language); I understand they chose the option that delivers the best IQ and widest DR (i.e. preserve as much from the highlights and shadows as you can)
Canon has a big advantage there because they have a software utility that anybody can use to define custom picture styles, including a luminance curve and color remapping

my own tests with my Canon 550D (T2i) are relatively close to their result (11 stops with picture styles available at the time they did their tests)
as I said before, my Flaat_3 and Flaat_4 picture styles bring back to life over a stop of previously crushed shadows (even with respect to, say Technicolor CineStyle), increasing the DR to 12 stops or a bit more, at the cost of noise (which may render some of this extra DR useless in some cases)