Calibrating for general use

questions about practical use of Neat Video, examples of use
Post Reply
smidoid
Posts: 9
Joined: Mon Jul 09, 2012 12:58 pm

Calibrating for general use

Post by smidoid »

I'm fairly new to Neat Video and as such I followed the tutorials with interest. They didn't help all that much - as the video I'd shot was largely unsuitable for calibration.

For those with a similar problem this solution worked well for me.

Use the tutorial videos certainly - but, and this the crucial bit - go out and shoot video specifically for calibration.

Set your ISO at max (6400, 12800, etc.) and shoot as much as you can at as many exposures as you can making sure you don't blow highlights OR drop into the shadow area. If you camera can show live curves, all the better.

It helps (if you can) to switch off anything like auto contrast on your camera which might harden the curves too much.

Shoot full frames of each exposure - at least highlight, mid tone (grey) and shadows making sure the subject is completely out of focus - you don't want any detail as that hides the noise.

When you get back, create and save your profile - even using several different shots - I used five in total - to create your profile and save it.

The results were astonishing; and while your mileage may vary, I hope this helps.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

The Neat Image Calibration Target can useful for such test shots.

Also, do not forget than different shooting modes (like ISO 6400 vs ISO 12800) can produce very different noise, so they require separate noise profiles.

Vlad
smidoid
Posts: 9
Joined: Mon Jul 09, 2012 12:58 pm

Post by smidoid »

NVTeam wrote:The Neat Image Calibration Target can useful for such test shots.

Also, do not forget than different shooting modes (like ISO 6400 vs ISO 12800) can produce very different noise, so they require separate noise profiles.

Vlad
Very true (I should have pointed that out) [slaps wrist]. It probably applies to different resolutions (1080, 720 etc.) too and may even vary at different frame rates over many cameras.

I've found it useful to name the profile with the ISO and camera: so 600D-6400-720 is a Canon 600/T3i at 6400iso, 720HD; 5DII-3200-1080 is the 5d mk 2 at 3200, Full HD and so on. Very handy if you're working with several cameras - and I expect, even the different builds of the same model.

Very strictly, we should probably note ambient temperature too - but when it gets down to that, it's probably reaching the limits of sensible video anyway. :lol:

I have to admit (and again, your mileage may vary) that although I've tried Neat's own target (displayed on an iPad) I got more predictable results from "real" video of real textures (cloud, asphalt, grass) etc.

Vlad and the team can probably explain why this is... I'm just reporting my finding.

The point is that using a target (either a paper one or some footage defocused in camera) allows you to get a terrific noise profile when you have some real edge-case footage as I did: but you must copy the footage settings to ensure that Neat can collect an accurate profile.

I've tried a bunch of noise reduction packages for video; and I'd gotten to a point where I was considering having to re-shoot a bunch of stuff with actors that are difficult to get. A bit like videoing a wedding or other one-time event and getting the dailies back covered in noise.

Neat Video saved the day, the project and probably my sanity.

Anyone reading this and still wondering if it's worth it - take it from me, it is absolutely worth it.

And you can take that to your Producer.
vvulture
Posts: 24
Joined: Sat Jul 23, 2011 5:51 am

Re: Calibrating for general use

Post by vvulture »

smidoid wrote: When you get back, create and save your profile - even using several different shots - I used five in total - to create your profile and save it.
Hi smidoid,
I need some advice and i was hoping you could help me.

I have a shitload of home videos i want to clean up... and i mean A LOT !! lol.
All of my tapes contain many different scenes ( shots ), ranging from dark indoor stuff to outdoor sunny. In building a profile, would it be possible to build one that would be "sufficient" to use on every tape ? They were all shot using the same camcorder with all settings identical.

Or do i need to separate all the scenes and build a different profile for each ?

What i am planning on doing is using the calibration target, and using the same camcorder, replicate the lighting conditions in order to build an accurate noise profile. Would this profile be suitable for all the scenes ( shots ) within each tape ?

I hope all this made sense..

Thx mate.. :D
smidoid
Posts: 9
Joined: Mon Jul 09, 2012 12:58 pm

Post by smidoid »

Short answer: in my experience, yes.

The noise effect is a function of many things - but its primarily caused by (a) sensor noise and (b) amplifier noise.

The nice thing about sensor and amp noise is that, for a given set of settings, they tend to remain fairly predictable - if that's the right word since noise, by definition, should be anything but.

The second feature of "digital" noise from a typical camera sensor is that the noise remains the same from frame to frame - which is why it's so bloody obvious. Noisy film (high-speed stuff) is less obvious because the grain structure changes 24 times per second with each frame, in effect, blurring the grain against a backdrop of action.

Given these points, NEAT is faced with the essentially the same noise profile for any given sensor and ISO (sensitivity). The amount of energy captured by each pixel does not determine how much unwanted noise it generates - and the upshot of that is this: if NEAT has a good enough profile, it should be able to clean up some pretty awful footage.

A lot depends on how much time you want to spend on each minute of video. Treasured shots might benefit from extra tweaking (particularly temporal tweaking which compares multiple frames) and fine tuning whereas less important ones would not.

The bottom line is, the better you can make that initial profile, the better NEAT video will be able to perform; but naturally, your mileage may vary.

I've tried the "other" noise reduction systems and I found NEAT was the best overall, so I wish you the best of luck.
vvulture
Posts: 24
Joined: Sat Jul 23, 2011 5:51 am

Post by vvulture »

Thx for the prompt reply.

Since my last post, i have filmed the calibrating target under several different lighting conditions, again ranging form dark indoor to sunny outdoor, using incandescent, fluoro and natural lighting for indoor conditions.

Now how exactly do i use this footage to create my profile ? How to i use all of them to create one generic profile ?

Cheers
smidoid
Posts: 9
Joined: Mon Jul 09, 2012 12:58 pm

Calibration

Post by smidoid »

This isn't obvious in NEAT - the Interface is a little weak in this regard.

The profile is additive.

You load each section (a good way is to load them in order of brightness, but it's not crucial) and then calibrate as you go. You can save your "progress" if you like at each step and then re-load it to go with the new footage.

Neat looks at Noise over a number of different levels from very dark to very bright - you'll see this explained in the tutorial video that Andrew Devis (IIRC) did to show how. I think this is the one.

http://library.creativecow.net/devis_an ... at-Video/1

It does take a little work to get it right, but once you do, the actual noise removal is very straightforward.
vvulture
Posts: 24
Joined: Sat Jul 23, 2011 5:51 am

Post by vvulture »

Again, thank you for your help..

When you say " You load each section ", do you mean each individual square on the calibration chart ? and click "manual fine tune " each time ?

Is it best to film the chart under low light to maximize noise ? I find that in bright light, noise is very minimal.

Thx 8)
smidoid
Posts: 9
Joined: Mon Jul 09, 2012 12:58 pm

Post by smidoid »

Hi VVulture, I'll explain in a bit more detail and hopefully this will make it clearer! :D

What you have to be careful of here is to note the ISO setting on the camera. It might be that your cam is trying to be "helpful" but adjusting the ISO (in effect, turning up the volume) when it detects a low-light situation. The problem with that is that it introduces a different noise source with a different profile into the mix.

There's not a great deal you can do about this with existing footage, but it's something to be aware of. I have calibration charts for the main (low-light) ISO settings we use on our films, in good light the amplifier and sensor noise is not a problem.

This, incidentally, has been a problem as long as cameras have been around - which is why Hollywood is where it is! The early cinematographers needed the most blinding sunlight they could get to effectively expose the early films.

Remember though, electronic noise is different to film because the grain does not change each frame.

The noise pattern from each pixel on the sensor should be broadly similar for a given ISO; but as the ISO (gain) rises, a second noise source comes into play and that's far more random.

TBH, I didn't use NEAT's calibration chart as it seems more suitable to stills.

I just fixed the ISO at 3200 (1600, 800 etc.) and shot a second of video of differently toned (dark, middle, light) subjects at a fixed shutter speed and aperture.

The key here is to ONLY change a single variable wherever possible: in this case, it's the tonal depth of the subject. Ideally, you're looking for grey tones - ones that have a good mix of red/green/blue - which is what the calibration chart should do.

So a section would be a second (in fact, one FRAME) of video of, say, some pavement or roadway; another might be a slated roof, another some lead flashing... and so on.

Dark colours are more likely to be affected since they reflect less energy (which we see as light) and that tends to confuse the sensor. You'll see this on the typical noise profiles as the noise curve tends to drop off from left to right.

To understand this, consider having to weight out some flour on a SINGLE scale. If the scale is calibrated from 1 to 1000 grammes for instance; and you need say, 700g, a 1% error amounts to 7 g. which isn't much; but trying to measure 5 grammes with a 1% error margin is a hell of a lot harder because you have to work within 0.05g on the same scale.

This is only a loose comparison, but it should give you some idea of why the noise gets worse in the darker areas where the measurement is more crucial. The measurement of collected "quanta" are drowned out by noise in the amplifier in the same way as you're more likely to make an error reading an imprecise scale when reading tiny amounts - which is what the dark areas are.

The problem gets worse as the amplifier has to work harder - hence noise increases as ISO increases.

When you consider all this it puts NEAT's job into some perspective and shows you just how damn clever it is.

A final thing to note here is that the nature of random noise is useful to us on very noisy footage. Each time the camera records a frame, the sensor is reset.

Compare this to the precise weight problem. The error created in the amplifier, being tiny but random, changes slightly with each frame. Just as you could make repeated measurements and, by chance, get an more accurate average value, NEAT can do this with the temporal setting.

Predictably perhaps, this extra work takes a long time - each subsequent frame has to be completely analyzed and averaged; but on very precious footage, it might be required to remove that last bit of crud.
vvulture
Posts: 24
Joined: Sat Jul 23, 2011 5:51 am

Post by vvulture »

You have been very helpfull my friend.. thank you :)

All my footage was shot using a miniDV Panasonic 3CCD Camcorder with all settings on auto, including auto-gain, auto W/B etc...

i'm getting a true sense of how all this works thanks to you ( and others ) and i have come to the conclusion that it is near impossible to create a single profile which can accommodate greatly varying scenes shot with the same camera, so what i have done is create a profile which gives me the best results under low light conditions. Brightly lit scenes don't really matter because noise is very minimal. I have also decided to increase my temporal radius to 3 and enable adaptive filtering.
Using these settings, it takes approx 2.5hrs to filter a 1.5hr clip.

Also, is it normal to get a very very small amount of colour shifting when a profile is applied ?

Cheers 8)
smidoid
Posts: 9
Joined: Mon Jul 09, 2012 12:58 pm

Post by smidoid »

You're quite correct: a single profile can't do everything with the current technology - if it could, NEAT would probably come with preset profiles! ;-)

The problem you have is that automatic ISO.

The shutter speed and aperture only determine the amount of light that hits the sensor (three in your case); the problems start at the sensor and amplifiers - so your problem is going to be dealing with the varying ISO - and that's going to be tricky given that you don't know what ISO the camera had selected for a given scene.

My Panasonic 3-CCD unit has a setting to fix the ISO (although it calls it "gain" which is technically probably more accurate).

It might be worthwhile calibrating a general profile, based on a high-amplifier gain and working from there. I don't envy you though - altering one variable (gain/ISO) is bad enough - add in others and it's going to get a whole lot harder!

Marc
Post Reply