Lamplighter55

On 03/12/2016 at 10:12 PM, Andrew Reid said:

I am slowly getting round to finishing my mammoth XC10 review.... expect it this week!

... in time for Christmas ... ;-)

17 minutes ago, Tim Sewell said:

Probably could use a bit more work - especially the score - but I quite like the 'thrillery' feel. Cinema EOS standard, HD, colour and contrast balanced then FilmConvert-ed to Fuji Velvia. Soundtrack done in Filmstro.

 

I think this is what we call 'challenging lighting conditions' - winter in the UK - grey day lighting. Nice grade - has a interesting rich textural quality to the colours. 

On 24/11/2016 at 4:46 AM, hyalinejim said:

This post is a bit dense, as I've been figuring out ghosting in relation to picture styles, ISO and 4K v HD so skip this if it's not of interest to you. If you are interested, don't take this as gospel - find what works for you!

I went and checked and it looks to me like EOS Standard has the least ghosting. Here's a motion test at ISO 4000, f5/6, 1/50s in 4K. I balanced the contrast between each of them and desaturated so that ghosting becomes more clearly visible and comparable. The area of grey in the middle would be completely clear of horizontal stripes if there were no ghosting in this camera. I'm counting (roughly) the number of times I see the white band repeating before it becomes so indistinguishable as to be unproblematic.

EOS STANDARD ghosts = 4

 

CINEMA EOS ghosts = 5 or 6

 

WIDE DR = 6 or 7

 

C-LOG = 8

Now, whether or not you see the same number of ghosts as me doesn't really matter. The general trend is clear and it looks like EOS Standard has the least ghosting.

This was at 4K and I tried the same test in HD at the same ISO of 4000. All four profiles were equally rubbish here but at lower ISOs the same trend emerged with EOS Standard having the least ghosting.

Interestingly, in 50fps 1/50s the ghosting appeared to be a little better than in 25fps 1/50s because the ghosts were smaller. So you might get better results by switching to 50 or 60fps, as long as you can live with the effectively decreased bitrate.

Also interestingly, in the EBU paper the guy talks about the resolution drop in HD at higher ISOs. I think Canon chose to address this in their last firmware update because now HD looks as if it might have more resolution than 4K at ISO 6400 - even when the 4K is downsized to match it!!! Check it out:

HD on left, downsized 4K on right:

In an individual frame, the HD looks much cleaner. But if I press play on this static scene it's pretty obvious that there is major noise suppression going on in the HD which feels a bit plasticky without added grain - and if there's movement in the scene then noise reduction and compression artifacts are everywhere. Still though, I'm very surprised that HD does so well - I can read the text much more easily than in the 4K. Perhaps the XC10 is now optimised for shooting charts from a tripod

OK, back to ghosting and low light. Since it was the best, I did some more tests using EOS Standard and for me around 1600 to 2000 is a good cut off point in both HD and 4k. Ghosting is beginning to appear at this stage but if you weren't consciously looking for it you probably wouldn't see it. Any higher than that in HD and ghosting gets progressively worse, while the noise stays constant and the image gets smeared. In 4K, on the other hand, ghosting seems to stay about the same as ISO increases, but noise kicks in so much that HD is actually better. So if I ever do need to go 3200 plus - to film the ghost of Elvis or something (as long as he doesn't move too much) - I think I'd use HD rather than 4K as there's less noise and the same or more detail. I was surprised to see today in my HD ISO5000 clip of the girls and skeletons how usable it looked. In 4K it would be a mess of colour noise at that high ISO.

So all in all, I'm fairly happy with EOS Standard as a workaround to mitigate ghosting. I can now utilise the ISO range from 160 to 1600 to help with exposure without worrying about messing up image quality too much. This captures a similar tonal range that you would get in C-Log at 500 - 5000... and that's a hell of a lot higher than I would ever have gone before. If I think back to the first video camera I ever owned, the Canon XM2, that had 3 stops of gain. Part of the reason that I like the XC10 so much is that it reminds me of that camera, shooting handheld when I was first discovering video. And I'm glad that I can now confidently use it in moderately low light.

 

 

 

The transform from 4K to HD (in camera) basically is a 2x2 sample of the 4K frame pixels going into one HD pixel - so effectively 1/4 the spread of noise. Essentially averaging to the mean each set of 4 pixels in the 4K to create one pixel in HD. The 'neat' thing about this is that you can (canon can) also use this to sharpen (increase luminance contrast) as part of the process as the chrominance noise is variable for each colour channel. So we end up with less noisy but slightly sharper images in HD.

I still think Canon should be able to improve on the 'ghosting' or 'temporal' quantisation/sampling problem - if they are not then they should be using a fast optical flow algorithm to integrate the frames at low frame rates - to get better filmic motion blur. As seen with our examples there is some kind of 'stepping' and 'kneeing' at the transitions from dark to light and visa versa - this is like the effect you get on CRTs when a high key signal drops to base black. Along the edge of the transition you see an over-shoot and undershoot of the image signal. Apart from the scan-line look,  this is what gives an old analogue video signal it's characteristic 'Video' look - odd dark lines around things. It is quite possible that the required level of processing is beyond the single Digic chip's spec for the XC10/15 and is only implemented in the higher end cameras (C100/C300/C500/C700 etc.) - can but hope otherwise.

5 hours ago, Lintelfilm said:

ARRRGH! More XC10 issues!

I've been using 100% zebras to help meter and of course guard against over exposure for a while now and it's been working great. However all of a sudden it seems to have stopped. Now, if I expose the sky in a landscape shot so that the zebras are a notch under 100% (i.e. no zebras showing but a slight tweak to up the exposure and they start showing), the sky gets completely blown - and I mean completely. No information at all! I've now tested this several times and the 100% zebras are clearly not accurate. Add on top of this there are supposed to be superwhites and I'm very concerned my unit has become defective. It's possible it was always like this but I'm pretty confident it wasn't as I've been exposing like this for a while now.

Can anyone contribute? How do your cameras respond to using 100% zeberas to expose to the right (just before clipping)?

Arghhh!

Humm could it be that the luminance gradient from the zebra cross-over values are much brighter than in previous shoots? Light being logarithmic the values could easily rise to a 'burn-out' level if you are shooting on a bright sunny day - for example any direct specular sunlight reflections will be orders of magnitude above 'correct exposure'. 

2 hours ago, hyalinejim said:

Interesting! According to this theory, would you expect ghosting to worsen with ISO, as we have seen?

 

Here's the latest update from Canon:
 

I guess you could interpret this in a few different ways. I'm going with "they're fixing the firmware" for now.

... the artefacts are quite characteristic of 'Temporal aliasing' but obviously some processing has been done as it does not seem to be global to each frame and seems to to be more pronounced when there is a pronounced difference in contrast from the preceding frame onto an area with less contrast. There should be a firmware adjustment that can solve the problem. (There's an interesting page on 'Red's site that helps explain temporal aliasing and their solution (hardware) in using an lcd layer between lens and capture chip that controls the light levels at a pixel level across the exposure of each frame.) It is a tricky problem though, that all digital recording devices have to deal with - the question for Canon is were in the image pipeline, can be mitigated.

Assuming this is indeed the problem then in theory we can reduce it by capturing a flat but brightly lit image (reducing contrast but also pushing more of the pixel values towards the mid-point exposure value), running at a higher shutter speed and keeping camera nodal movement to a minimum (to reduce spatial aliasing). I'll also try some tests to see if I'm correct - at least in part ;-)

[Link to Temporal Aliasing on Red's site: http://www.red.com/learn/red-101/cinema-temporal-aliasing]

2 hours ago, Lamplighter55 said:

.. and from Wikipedia 'The XC10 uses a single DIGIC DV 5, while the C300 Mark II uses a dual DIGIC DV 5 implementation.' So less sampling/super-sampling no doubt. There is also now the DIGIC DV 5+ (faster by 3x - so possibly equivalent of the dual set up in the 300 MkII) - yet to find which camera platforms it's to be used for apart from the EOS 1D X ... maybe XC15?!  

The technical term for these artefacts is 'Quantisation error'. Which could also hint at a way to mitigate them by spreading the pixel values over a wider range of bits - in other words shifting the luma and contrast with more light, assuming under really low light (higher ISO settings) the quantisation is lost/masked in the higher digital noise signal.

On 01/11/2016 at 9:44 AM, hyalinejim said:

I agree. All of the seven or eight XC10 ghosting tests that I've seen have shown it, but it doesn't mean all XC10s have the problem. If anyone has a clean XC10 please post footage.

And the one clean XC15 test doesn't mean all XC15s are free from the problem. If anyone has a ghosting XC15 please post.

If you're thinking about getting either model I would strongly recommend you try before you buy, or look into your seller's return policy.

I bought my XC10 back in January (2016) and have been very happy with the quality of the images/footage. This 'Ghosting' or temporal noise reduction (possibly) has also shown up on my footage (see bellow) - and I've been following this thread on the issue. Having had a close look at the artefact I can pretty much say with certainty its an error in the calculation of how Canon integrate values from one frame to the next when calculating for motion blur. Basically the maths is off and creating a 'rounding error' so the pixel bins either hold an 'over shoot' or 'under shoot' value - so you get this characteristic 'zoning' of luminance values. Its a bit like the posterising effect (thresholding) one can achieve when reducing the bit depth of an image. So what is causing this ... hopefully it's a firmware only issue, in which case the frame to frame integration algorithm needs re-writing, or it is something done in hardware ... which means a fault in the DIGIC DV 5 chip. Canon will need to check in the later case if it is a batch issue or systemic. Let's hope it only requires a firmware update! The 'good' thing is that if it is (as I surmise), then the solution is just a question of maths - but the problem is how that is implemented - at chip level in hardware or a firmware 'bug'.

(I should add, this could also will explain why this is less evident on HD footage as it has 4x more 'integration' as a reduction from 4K in spatial resolution - equivalent of averaging down to a quarter sized image.)

One test we could try is running at a higher frame rate (HD only) and seeing if the artefacts are reduced or increased further for the same amount if spatial moment - in other words more samples over a given time frame and spatial transfer.

I recently bought an XC10 having researched the pros and cons and also taking advantage of the 'January sales'..  So far am pretty happy getting used to its capabilities. I have a question ... I have been trying to find out if any of the 3rd party 1D 'cube' LUTs available for the C100*/300* are compatible with the Clog setting of the XC10? Anyone edited with these in FCPX for example? They seem to work but I'd prefer to be certain, as when it comes to 'pushing' the limits when capturing I'd like to know I can get a safe recovery of the footage with out bias, once I've transcoded it for editing. So any ideas or feedback most welcome. Thanks!