Does this noise look right?
Posted 10 May 2007 - 04:19 PM
We're wondering if the noise we're seeing in our F900 footage is normal. It was brought into our Avid at 1:1 and exported as uncompressed tiff sequences. There are some interesting horizontal lines (especially obvious in the blue channel) and more noise in general than the vfx supe saw on the monitor on-set. Just wondering if anyone can take a look and give me your opinions. I uploaded a 20MB uncompressed Quicktime example here (a small crop of the 1920x1080 at 100% scale):
Let me know if that looks fine to you, or if we need to investigate. I can't show any of the actors footage, but it's very present in the mid-tones. Is there any step in the Avid ingest that could be contributing?
Thanks for any help!
Posted 10 May 2007 - 07:38 PM
Posted 10 May 2007 - 08:21 PM
Posted 10 May 2007 - 08:32 PM
Posted 11 May 2007 - 01:22 PM
Posted 17 May 2007 - 09:15 AM
There is definitely visible noise in the blue channel of the Cinealta.
The Varicam F did not hold up in my opinion either. They are fairly old cameras now I suppose.
Posted 18 May 2007 - 06:36 AM
A Preliminary Objective latitude test that am running with Cinealta F900R and Varicam F the last days for evaluating the cameras in order to calibrate them in order to get the maximum performance the preliminary tests show the following:
The set up is using Danes Picta transmitive step chart with 28 steps of 0,15 density this equals to 1 stop on every two steps a total of 14 stops when the chart is backlight.
A grab from the HDSDI out to an FCP in uncompressed 10bit convert a still frame to tiff with combustion and then import it to imatest for noise to latitude measurement. More info in imatest web site
The exaggerated (+33% brightness and +77% Contrast) tiffs with Photoshop to show the effective visible steps are following.
The noise analysis is following.
Note that Cinealta was on Hyper Gamma 3 and -3db after a factory settings reset with a black balance and a preset white balance in tungsten. Similar thinks with Varicam F Film REC, 0,45 gamma, -3db, 500% DL and black balance with preset white in tungsten after a factory reset.
What is obvious, is the problem with the noise in cinealta and the lack of good quality latitude even with no tape intervention.
I don?t know yet what I can do to make F900R to perform better. I?m on the search.
What is clear to me is that Varicam is far better than cinealta on the latitude/noise aspect with more than 2,7 stops diference in high quality (very low noise) and 1,87 in the medium renge.
2,7 stops is like a car is going with 50 miles/hour and the other one goes with 360 miles/hour, so the difference is huge!
For further evaluation the original files unprocessed are in original files 1,8Mb
Posted 18 May 2007 - 08:31 AM
I apologize for not being able to give specific numbers and examples of the test, but it was not done for me and I no longer have access to the footage.
Posted 21 May 2007 - 02:22 AM
Do any of you recognize that particular horizontal line noise that's in our footage? I don't know if you've been able to download that cropped clip and look at it, but that noise is pretty blatant and is just a strange pattern in my opinion. We're having to roto some otherwise-well-shot bluescreens because of noisy jaggy edges.
I don't know if it's appropriate to ask about other cameras in this forum, but we're trying to learn from this footage so we know what to plan for in future shoots. If you got footage back like that, would you say "yeah, looks about right for the F900" or would you investigate the problem? lens? setup? chips? enhancement? tape deck?
Posted 27 June 2007 - 05:34 PM
I fear your test is flawed.
You took frames directly from the HDSDI output from the f900R camera?
This output is not full specification, and according to Sony and others it is noisier and as the tech manual says is for monitoring only.
(I know the ad men say you can digitise from the camera, well you can but it is not ideal quality, usual Sony brouchure misinformation )
I know of one other instance in the last month where this has caused confusion.
A Miranda adaptor or Sony HDSI brick adapter will produce "full quality" HDSDI from the camera
Replay from the camera is to be avoided for digitising.
Best test of (any) camcorder is to record then playback in a deck to a NLE unless you are using the camera as a feeder.
Try Digital Praxis curves as anything from Sony with a "Hyper" should be used with caution. If in doubt use No 6.
Posted 27 June 2007 - 05:46 PM
ask Panavision what software version is in the camera.
Early f900s had diagonal lines in varying forms which were fixed with software and hardware upgrades.
Was the camera connected to mains power? In an older building?
Could a earht loop have been created beteen camera monitor or sound department?
Was the camera shipped to you on Fedex UPS ect
This kind of handling often lossens a board whic can create noise in the blacks.
Posted 04 July 2007 - 03:51 PM
Yesterday I was testing again the F900R and I see what Jance is describing in a certain level of exposure a horizontal noise pattern that was vanishing when light was a little up or down by ¼ of a stop.
As for the output of camera I haven?t found (even in op. manual) anything saying that the HDSDI out is inferior of what is being recorded on tape.
Actually this is compliment the camera since in that way we have 10bit 4:2:2 uncompressed recording in full bandwidth and not HDCAM compression in 1440x1080 3:1:1 8bit.
Lastly it would needed extra electronics to downgrade a signal that comes from camera head and this is not cost effective for a manufacture.
If you have any documentation about your argument I would like you to make it public. How I can get the best curve for Cinealta? We have an official test in two weeks? can you point to me the best settings that you have to surpass the performance that I measure?
Posted 04 July 2007 - 04:06 PM
> Lastly it would needed extra electronics to downgrade a signal that comes from camera head and this is not cost effective for a manufacture.
Hypothetically, if they were doing the 1440px downsample before the YUV conversion, which they probably are as it would reduce the DSP bandwidth, you would expect to be monitoring the 1440px image as they'd otherwise have to include a second YUV converter.
I don't have sufficient time on the F900 to answer these questions directly, but I can offer a few principles:
- Is it possible that the monitorinig output on the F900 is post compression, or at least post 3:1:1 subsampling, or post HDCAM downsampling? That will, fairly obviously, cause the noise to appear horizontally elongated. Compare the same frame from tape and from the output (obviously, capture the tape image on a proper deck).
- The green and magenta fringing is an optical artifact. I'd estimate in this case that this section was taken from the upper left corner of the frame, but depending where the focus was set I could be 180 degrees out. Depending on the lens setup, this could be viewed as quite bad. Where'd you get your glass, Coca Cola?
- All CCD cameras are noisiest in the blue on the basis of nothing more than the optical physics of silicon photodiodes. This has been reported in F900 before.
Posted 05 July 2007 - 08:16 AM
As Phil said colour fringing on left hand side of frame is a characteristic of many opticval blocks.
I did a comparison for High Definition magazine it may be downloadable form the www.definitionmagazine.com
It has frames from Fuji Super zooms and Canon zooms that both suffer the fringing.
In respect to f900R output the best approach is to use a seperate HDSDI oytput and compare it to onboard output.
Sony and those who have the new HDC F23 camea for rent would prefer us to use the F23 to record into SRW1.
So any fact/rumour/hint/disinformation they propogate to encourge use of F23/SRW1 combo is to their adavntage and
helps them sell/rent F23 cameras.
I don't know if f900R on board output is inferior to Miranda MDC/evertz adapter or not as I haven't done any tests.
However and respected en
Posted 05 July 2007 - 01:07 PM
Per the color fringing, that example image I posted was indeed from the upper left of the frame. The laptop was sideways inside a gimbal room that rotated.
Unfortunately I'm not that much of a camera or on-set guy; I'm coming from the post side and being asked to help diagnose visual artifacts after the fact. (Currently dealing with bad vertical banding from a Phantom camera... grrrr).
I'm personally just pushing for our shoots to have someone look at frames under a microscope as soon as possible, on-set. Correct me if I'm wrong, but that particular noise pattern wouldn't really show up on a scope, or at least not as clearly as looking at separate channels at 100% zoom. I mean, the first thing I do as a compositor is zoom in and look at R, G, and B separately, and usually my reaction is "oh poop."
I'm really looking forward to some of the new cameras and new workflows. I won't go so far as to say film is dead, but I personally would be glad for tape to die.