D5200 Anyone got it ???
- Thread starter ChipThome
- Start date
homestar_kevin
Well-known member
This is getting sweeter by the day.
I'm very intrigued by all this, want to see it to a ninja as well
I'm very intrigued by all this, want to see it to a ninja as well
I downloaded the original 1280x720 MP4 from your Vimeo link and examined it on my calibrated grading monitor. The limited 720p resolution of the video makes the coarse quantization of graded up shadow details much more noticeable than it is in JPEG stills. In addition, low-bitrate H.264 encoding degrades shadow details more harshly than high quality JPEG encoding. These factors combined can readily account for the image quality differences between H.264 video frames and JPEG photos straight from the camera.
I don't see any evidence in these sample videos that justifies your speculation that the D800 (or D5200) debayers its RAW sensor image "to an uncompressed 8-bit 422 image (my guess), and then the picture control is applied". Regardless of how you manipulate the in-camera gamma curve, with Flaat_13p or any other profile, the darkest details will inevitably be coarsely compressed by the encoder. This isn't usually noticeable since those details are very close to black anyway. But if you dramatically boost the shadows into gray in post-production, the coarsely encoded image quality is immediately apparent, especially in video where the resolution is relatively low. It's not due to engineering incompetence on Nikon's part; it's an inherent limitation of the 8-bit encoders used to compress the image.
8-bit scaling is in general a wretched way to capture shadow details. Here's a link to a short video where Charles Poynton describes the notorious 8-bit "code level 100" problem in detail:
http://www.fxphd.com/blog/this-week-in-fxphd-color-science/
Last edited:
squig
Veteran
Samuel H
Major Contributor
I'm not dramatically boosting anything in post-production. Those images are straight out of the camera, both for the jpg still and for the video clip, and yes they are that different out of the camera. And I'm not accusing Nikon of incompetence, there may be a very good reason for doing what they do. I'm just observing what they do and how it affects my ability to get a nice image out of the camera.
Please stop it, or run your own tests, or go on thinking whatever you like to think. Thanks. http://xkcd.com/386/
Last edited:
ReelFilmGeek
Well-known member
Quick question if anyone would know, should off brand batteries from the d5100 work with the D5200 and what about battery grips?
All right, I tried being respectful, and posted photographic evidence and detailed explanations that refute your unfounded personal speculations on Nikon's image processing technology. Your own evidence is inconclusive, and unless you're a DSP engineer with access to Nikon's proprietary design documents, there's no way you could authoritatively make these specific technical claims as objective facts. I think it's an important function of forums like this to publicly pin down the kind of rumors you've been propagating, and determine whether there is any substantial evidence to support them.
John Caballero
Veteran
http://www.amazon.com/gp/product/B005DP9I14/ref=oh_details_o01_s00_i00?ie=UTF8&psc=1
http://www.amazon.com/gp/product/B004VHULS4/ref=oh_details_o00_s00_i00?ie=UTF8&psc=1
The off brand battery only works when you insert it in the grip after an original NIKON battery is already in.
Samuel H
Major Contributor
You're completely right. But I feel you didn't read my posts, that's why the discussion is not very productive. I say "the camera is cutting corners when debayering for video" and you bring stills as proof that it is not so. Then I post a still and a short video saying they're right out of the camera and "with exactly the same settings and picture profile, and not processed any further", you say the pattern may appear because I'm pushing my images up.
I've spent some time trying to understand how the camera works, and that's the conclusion I've reached. It may be wrong. Feel free to run any tests you like and reach whichever conclusion you end up reaching.
It's not that important, really...
Last edited:
BobbyMurcerFan
Well-known member
Samuel, I have the utmost respect for both you and Lpowell. I may have one possible explanation. It seems that the externally recorded video you posted was shot at 720P (b/c the recorder would not work at 1080P). Is it possible that the crappy quality you documented was actually due to a crappy downconversion to 720P? Thanks!
An h264 encoder will drop those bits down if the bitrate is low so a 16-bit encoding wouldn't help at all if the bitrates stay low. Mostly H.264 actually drops the 8-bits down to 6-7 bits in most scenes internally. Darkest details are not necessarily always coarsely compressed. There is a HUGE difference in shadow quality when using the FS700, encoding with AVCHD and going from 0db gain to +18. And it has nothing to do with the encoding as it encodes everything beautifully (even the shadows) when the gain/iso is down.
You can capture clean 8-bit material straight out of the HDMI. But that doesn't seem to be the case with the D800 and it does seem that the picture profiles are applied after the image has been debayered and in 8-bit space. That accounts for the banding.
I don't know why you guys are discussing only shadows because this should be visible everywhere. It could be resolved by just uploading a couple of shots with D800, one without any profile adjustments and one with. Preferably a noisefree scene and not-to-vimeo (that means, hosting the actual files somewhere)
Last edited:
Whaat? That vimeo file looked horrible, wasn't even in semi-HD. People blame the cameras for all kinds of things nowadays.
squig
Veteran
Did you download it?
I can't be sure of anything till mine arrives but it wasn't pretty.
Samuel H
Major Contributor
Yes, it could be. But I'm seeing that heavily posterized noise in the shadows in anything shot with the most extreme Flaat profiles, not just that clip.
I would love it if I could get, when shooting video with a very aggressive profile, the same shadow detail of jpeg stills shot with that same profile. But it seems not to be the case, to me it seems that for stills the camera works on a high-bit-depth source, and for video it is a low-bit-depth source. My guess is that it is the debayering that is creating the issue. It could also be sensor readout: the CMV12000 can do 150 fps when using 10-bit ADC but only 90 fps when using 12-bit ADC, maybe the D800 needs to use a lower-depth ADC to get 30fps out of the sensor (edit: this would have a disproportionate effect on the shadows, but I don't think it would look like what the footage shows). Whichever it is, they're cutting corners somewhere, and the effect is the same: heavily posterized noise in the shadows.
I'll be VERY happy if the D5200 doesn't have this issue (which could be the case, for example, if the issue is because of a lower-bit ADC for video, because this is sensor-specific and this is a different sensor, or if the pixel binning is done directly at the sensor -which is probably the case- because then the debayering algorithm has to work with a somewhat less data -compared to line-skipping- and may be able to use the high-quality option)
Last edited:
BobbyMurcerFan
Well-known member
Whatever the cause, I think there is reason to be somewhat optimistic about the D5200. The 5200 has worse rolling shutter than the 800, which might indicate that the sensor is being read more fully. I was always impressed with the minimal rolling shutter in Nikon's recent crop of cameras, but maybe they took some shortcuts in processing to get there.
OK, that's a valid point, so I shot a comparison in video mode as well. These clips were taken on the D5100 in 1080p24 mode at ISO 800. As with my previous test, I used a Custom Profile I made out of the built-in Neutral Profile, modifying it with a curve that boosts shadow detail. I imported the MOV files into After Effects, exported still frames into Photoshop, and cropped them at 200% magnification. If, as you've claimed, the D5100 uses coarse 8-bit precision when applying Custom Profiles to a video, we should see a significant degradation in the Custom Profile image quality compared to the built-in Neutral Profile.
For the first comparison, I graded down the Custom image frame with an inverse curve in Photoshop, matching its illumination to the ungraded Neutral image frame:
Custom Profile - Graded down @ 200% Magnification:
Neutral Profile - Ungraded @ 200% Magnification:
I did the opposite for the next comparison, grading up the Neutral image frame to match the illumination of the ungraded Custom image frame:
Custom Profile - Ungraded @ 200% Magnification:
Neutral Profile - Graded Up @ 200% Magnification:
Aside from minor differences in grading, I see no significant degradation in image quality among these video still frames. This would not be the case if the camera were restricted to using 8-bit precision to map the pixels from the built-in Neutral Profile into the Custom Profile's gamma curve - we'd see coarse banding in shadow details where there were 8-bit round-off errors in the remapped pixel values. The absence of these kind of artifacts indicates that Nikon does use high precision math in the camera's internal image processing.
In my view, these results should serve to vindicate the work you've put into your Flaat Custom Profiles. Tests like these demonstrate that the use of Custom Profiles does not produce perceptible image degradation, either in still-frame JPEG images or in H.264 videos.
Last edited:
Zephyrnoid
Veteran
I really hate double checking your hype Andrew....
http://www.eoshd.com/comments/topic/2125-exploring-nikon-d5200-hdmi-output-review-update/?p=27650
ReelFilmGeek
Well-known member
I should expect my camera to ship in tomorrow if everything goes right and will begin messing around with it. Sadly I don't have a hyperdeck to test external recording to though. Is there any place to read up on using nikons for video as I am use to using canon DSLRs with magic lantern.