Internal SDHC Slog Vs. Cinegamma 4 grading test + original .MTS files

[PrimeHD]

From what I understand as latest developments at CD, they figured Sony's 2K RAW is not good enough except for HFR (high frame rate) stream capture, due to Sony's way of pixel binning. So CD is trying to get around the pixel binning by capturing 4K RAW bitstream, and deriving 2K and HD out of it, AFAIK.

Question is, will CD's Odyssey 7Q debayer and store those files as compressed video - say DNxHD or ProRes, or will it store it as RAW DNG?

(Clearly we could do the conversion in post, since Odyssey 7Q also captures 4K RAW into DPX file sequence. HOWEVER, that would be very costly to us the users, both time-wise and in terms of Odyssey SSDs - of which TWO will be required to capture any 4K RAW stream. Conversely, IF Odyssey 7Q actually does better-than-Sony-quality downconversion/debayering internally, and records a resulting compressed HD video, then only ONE SSD will be needed, and no post-production time. 10 bit ProRes or DNxHD, with 4:2:2 chroma subsampling, is my current preference for sure, since I don't need 4K most of the time!)

 

[SKYLIKE]

U are asking the same questions over and over again PrimeHD.
Odyssey7Q wont capture 4k raw, just 4k yuv. 2k and HD are derived from 4k and will be stored in compressed formats. 2k also available as 2k RAW. Debayering takes place everytime a compressed/uncompressed Format is stored because the FS700 only outputs RAW.
Visit CD website. There are all formats explained.

 

[jcs]

2K/HD are available as uncompressed RGB/YCrCb 12-bit, which if done correctly will provide full resolution and no loss of color detail vs. RAW, making 2K RAW, with lower resolution after deBayering, not worth using.

 

[cuervo]

2K/HD are available as uncompressed RGB/YCrCb 12-bit, which if done correctly will provide full resolution and no loss of color detail vs. RAW, making 2K RAW, with lower resolution after deBayering, not worth using.

Right on!!

 

[Razz16mm]

2K/HD are available as uncompressed RGB/YCrCb 12-bit, which if done correctly will provide full resolution and no loss of color detail vs. RAW, making 2K RAW, with lower resolution after deBayering, not worth using.

If this is truly uncompressed 12 bit RGB without any chroma subsampling, the file size will be 3 times larger than DNG Bayer raw too. But I think Sony is playing games here and their DNG raw mode is not literally a pixel for pixel raw luminance data dump. They are not big fans of Bayer raw acquisition.

This is what a native 2k pixel for pixel DNG raw frame can look like. Shot at ISO 400 and pushed to ISO 2600 with no noise reduction and a small amount of sharpening added in the Debayer process to a 16 bit uncompressed TIF RGB which Flicker compressed to 8 bit JPEG.

. Show me a single 2k RGB or DNG raw frame processed out of a Sony with comparable quality.

 

[paulcurtis]

Show me a single 2k RGB or DNG raw frame processed out of a Sony with comparable quality.

You serious or just trolling?

There's nothing special about that frame at all, it was shot with the bolex thing wasn't it? If memory serves it was also shot with a classic canon rangefinder lens? Hardly a demanding window to test a camera? (I have the canon f1.2 version, beautifully flawed lens)

Most cameras out there can shoot something that looks like this. What about demanding scenes with a large range of light, or very low light?

At least the sony test footage with the AXR was harsh daylight and there was some great latitude in those files (the japanese women dancing). Or if you take a loot at the 4k dpx's Peter posted on his thread. I'd like to see what a CCD sensor would do with the highlights on that...

cheers
Paul

 

[Razz16mm]

You serious or just trolling?

There's nothing special about that frame at all, it was shot with the bolex thing wasn't it? If memory serves it was also shot with a classic canon rangefinder lens? Hardly a demanding window to test a camera? (I have the canon f1.2 version, beautifully flawed lens)

Most cameras out there can shoot something that looks like this. What about demanding scenes with a large range of light, or very low light?

At least the sony test footage with the AXR was harsh daylight and there was some great latitude in those files (the japanese women dancing). Or if you take a loot at the 4k dpx's Peter posted on his thread. I'd like to see what a CCD sensor would do with the highlights on that...

cheers
Paul

No, I would just like to see some full resolution stills frames without web video compression. There is more to it than just DR.

I have been playing with the original FS700 MTS files posted by the OP. They are quite soft and a bit noisy by comparison. Just like to see what these cameras can potentially do at their best with 12 bit RGB or Raw at 2k.

 

[Razz16mm]

You serious or just trolling?

There's nothing special about that frame at all, it was shot with the bolex thing wasn't it? If memory serves it was also shot with a classic canon rangefinder lens? Hardly a demanding window to test a camera? (I have the canon f1.2 version, beautifully flawed lens)

Most cameras out there can shoot something that looks like this. What about demanding scenes with a large range of light, or very low light?

At least the sony test footage with the AXR was harsh daylight and there was some great latitude in those files (the japanese women dancing). Or if you take a loot at the 4k dpx's Peter posted on his thread. I'd like to see what a CCD sensor would do with the highlights on that...

cheers
Paul

No, I would just like to see some 2k resolution stills frames without web video compression. There is more to it than just DR.

I have been playing with the original FS700 MTS files posted by the OP. They are quite soft and a bit noisy by comparison. Just like to see what these cameras can potentially do at their best with 12 bit RGB or Raw at 2k outside of a conventional video environment.

Sony's claiming up to 14 stops DR for these cameras, yet the base SN ratio listed on their spec sheet is 57dB. The D16 claims 12 stops with a base SN ratio of 64dB. Where is the range coming from and how are they getting there? What does it cost in terms of image quality in other areas? There is no free lunch.

 

[cuervo]

here's a couple of shots made with Slog2, on my FS700, and captured to a Pix240 recorder(8-bit, 422, 1080p24). The footage was ingested to DaVinci Resolve where an S curve was applied, in addition to saturation and a little bit of sharpening....take note of the small hairs on the stem of the plant shots, and the feathers on the hummingbird.This web image isn't nearly as distinctly clear as my originals when viewed on my LCD...oh well.

 

[jcs]

Show me a single 2k RGB or DNG raw frame processed out of a Sony with comparable quality.

How about a 24Mbps AVCHD/H.264 frame (size reduced by this forum)?

View attachment Route-66_1920.zip (full resolution JPG).


2K RAW can't resolve even 1920 horizontal resolution after deBayering. The image you posted looks nice, but does not have a lot of resolution (looks kind of like 5D3 H.264 with a little post sharpening, which is around 1620 resolvable pixels. A resolution list I created a while back: http://www.eoshd.com/comments/topic/1953-comparing-actual-camera-resolutions/). 5D3 1920x1080 RAW looks really nice after ACR deBayering, NR, and sharpening, however the FS700 24Mbps AVCHD still provides more resolution.

 

[Razz16mm]

Nice looking image. But no it doesn't resolve any better than a high quality debayer of a 1080p raw frame. No fixed digital image format can optically resolve an image equal to its pixel resolution, even in monochrome.

Fact is that all of these new cameras are reaching the upper limits of practical resolution for the HD format. And that is around 1.6k horizontal unless you want to put up with significant aliasing artifacts. The Alexa shows an MTF of only 10% at 1.6k, virtually no contrast, so that is its effective upper resolution limit in HD too. If you want that smooth nicely textured film like image quality 1.6k is about as good as it gets.

The reason low bitrate AVC out of Bayer video cameras doesn't resolve that well is due to the need for lower quality debayer algorithms that can run at video frame rates real time.

But the reason one would choose to shoot raw is precisely to have the far greater range of image control that you have with sophisticated raw processors. The below image is a day for night grade of the same Bolex frame. With a good debayer algorithm one can retain virtually all of the luminance resolution of the full color image, something that is not possible with subsampled high compression consumer grade video codecs like low bit rate AVCHD. In this image the full resolution of all three color channels was applied to the blue channel. Current high quality debayer processors can like the human eye process the full luminance resolution of the sensor separately from color resolution using every pixel of the sensor regardless of what color filter is in front of it.

Try this with a low bit rate 4:2:0 AVC frame and see what happens. AVC is a fine delivery codec, but a poor acquisition codec unless you are satisfied with the image as it comes from the camera.

 

[jcs]

Nice looking image. But no it doesn't resolve any better than a high quality debayer of a 1080p raw frame. No fixed digital image format can optically resolve an image equal to its pixel resolution, even in monochrome.
Try this with a low bit rate 4:2:0 AVC frame and see what happens. AVC is a fine delivery codec, but a poor acquisition codec unless you are satisfied with the image as it comes from the camera.

The FS700 is clearly resolving more resolution than the ~2K RAW image posted. Even in 420 AVCHD (which has full resolution in luma). We need a 4K sensor to sample 2K without aliasing (resulting in 1K lines pairs per resolution charts); that's Nyquist sampling theory. Here are some FS700 charts: http://provideocoalition.com/awilt/story/high_speed_and_low_light_with_the_nex-fs700/. There are more out there (trumpets, etc.). If you can provide a test chart for the 2K raw camera then we can compare apples to apples. That'a why the C300 does so well resolution-wise- 4K sensor averaging green to get 2K (HD) and using R & B as is (outputting 422).

Regarding the ARRI crapping out near 1.6K (10% MTF)- that makes sense from a 3.5K sensor: 3.5K/2 = 1.75K, just as we'd expect per Nyquist. However, in this case the drop in detail/smoothing is from their OLPF.
DeBayering quality varies from worst case: 1/2 sensor resolution (fast) to closer to 100% sensor resolution (very expensive to compute- similar to your blue image example (using all values to compute luma), but applied to a full color image). ACR does a really good job- the 1920x1080p 5D3 RAW looks really good (but not full 1920x1080 resolution (not "960+ line pairs on a test chart")). The FS700 provides higher resolution (and the C300 provides pretty close to the theoretical limit). ARRI rocks because they provide high enough resolution, and completely nuke any aliasing/artifacting (very trick sensor and optics; from studying film transfer to digital before building a camera).

 

[jcs]

Side by side is easier to compare:

Full resolution: View attachment FS7002KRaw.zip

FS700's 4K sensor sampled down to HD is providing more resolution than the 2K RAW image (while possible to get more resolution from the 2K RAW image, once moving as video, aliasing can be a problem if 2K is the native sensor resolution).

 

[paulcurtis]

Try this with a low bit rate 4:2:0 AVC frame and see what happens. AVC is a fine delivery codec, but a poor acquisition codec unless you are satisfied with the image as it comes from the camera.

You're preaching to the converted, most folk on here would agree with that and no one is really saying any different. Although the image you posted shows nothing taxing, even for AVCHD. Of course you're posting it as a compressed sample anyway, so any subtlety you may be seeing has long gone by the time it gets to us.

I'm not quite sure what you want to get out of this discussion? Download the 4k DPXs that Peter kindly uploaded and that will give you a good idea, although i am questioning the exposure on them a little but they're lovely examples - especially with the fire, notorious for being difficult to work with.

You can't possible be comparing the bolex to the 4K 700 solution? I personally made CCD based cameras out of various bits and pieces when everyone was doing the same. CCDs are very nice, produced (at the time) better images than CMOS but suffered badly in extreme ranges with highlight handling. CMOS sensors are several generations ahead now.

Sony isn't quoting 14 stops on the 700 AFAIK, i don't think they've quoted anything and initial testing has indicated 12 to 12.5ish.

If you *are* comparing the bolex to the 700 4K, you've got probably the same 1" kodak sensor i was playing with compared to S35, And you're getting 2K debayered image which colour sampling wise wouldn't be anywhere near close to 4k -> 2K sample. I don't see a valid comparison.

Let us know what you're trying to get to with the comparison?

cheers
Paul

 

[paulcurtis]

Side by side is easier to compare:
FS700's 4K sensor sampled down to HD is providing more resolution than the 2K RAW image (while possible to get more resolution from the 2K RAW image, once moving as video, aliasing can be a problem if 2K is the native sensor resolution).

It's also pretty difficult to compare skin tones when one is lit with green! I don't see what the OP is really seeing with the bolex one that is 'special', i think even the guys themselves said that it was done pretty ad hoc.

A more valid comparison would be with the pocket camera from BM.

I would say though that the 4K - 1080p scale within the 700 is flawed and not perfect. I think when you can see a post version of 4K - 1080p by comparison you'll realise that... Or take a still with the 700 (4K) and downsample that and side by side it with the AVC. Night and Day.

In fact if the OP wants to see what can be done, even in JPEG, shoot a still...


cheers
Paul