chromakey tests, anyone ?

[ulisses]

hello,

I'm curious about chroma key tests with the 4:2:0 footage.

thanks,
Ulisses

[matthew77]

Yes - much better than I expected - surprisingly smooth.

I can't post footage because of client/copyright issues but I'll try to post a still.

One thing I noticed was the difference between key frames and derived frames. There key frames are very smooth and the derived frames are a bit blockier. In motion it's invisible, but this is from frame by frame inspection. This is in the out-of-focus areas, in the gradient that I see the blockiness.

On sharp edges there is no such blockiness. The keyer I used was Keylight and it did a phenomenal job, with one click. No twiddling at all.

[Lenilenapi]

There are some shots taken outside in front of a green screen on DVInfo net though it was posted maybe before Christmas in the EX thread. Looked pretty good with hardly a complex lighting setup. The person who shot them was very happy with the result.

You could probably find them with a search on green screen.

Lenny Levy

[ulisses]

found some theads:

http://www.dvinfo.net/conf/showthread.php?t=110642&highlight=chromakey

http://www.dvinfo.net/conf/showthread.php?t=108630&highlight=chromakey



thanks !

Ulisses

[Lenilenapi]

Thanks Ulisses that's what I was talking about.
Do those look convincing to you guys ?
I'm no expert on this, but I do need to be able to tell my clients whether they will get a decent Key.

The HVX was just on the cusp. Do-able but we needed to be at 1080 and upped the green to 55IRE to deal with noise.

[matthew77]

The guy in front of the building was a bit hard-edged. He cranked up the contrast of his matte too much. He complained about motion blur and said that next time he would crank up the shutter speed. This is the wrong approach.

Proper rendering of motion blur and defocus blur is essential to a convincing key, because these are natural aspects of the footage. With Keylight and the EX1 I was able to get some pretty amazing (very) soft edge mattes that only exhibited blockiness in the gradients, and only really visible in still frames.

[Stevet]

If you're real concerned go with the SDI 4:2:2 out.

[Jim Arthurs]

I was going to start a new thread, but saw this one... here you go... comparing the native codec vrs. the HD-SDI out...

I find that the native 35mb/sec codec keys as well as any HDCAM 3:1:1 source I've used, and the HD-SDI out to uncompressed is just as nice and tight as an F900 out "live" to a single 4:2:2 channel of an SR deck.

Blue screen source clip, recorded internally and also to 8bit uncompressed at the same time. 3000K lights (the worst possible case for noise... blue screen and overly warm tungsten. It only gets better with green and daylight balance, BTW.)

The XDCAM codec side of the test clip had a slight chroma blur applied while exporting from Vegas (Chroma blur .3 by .3 is enough). -3db on gain. The lens was wide open if I remember...

Right click, "save as"... 25megs

http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_compressed_vrs_uncomp.mov

The main pointer I can give is to completely turn off detail. In fact, turn it off for anything you'd ever shoot unless someone ASKS you for that look. You will loose no real detail, as you can regain everything back with post sharpening. I'm surprised I haven't seen more clips with detail off... it was literally the first thing I did after testing to make sure I wasn't giving up any resolution.

Regards,

[matthew77]

jim,

Good point. I always shoot with detail off. Otherwise you'll get a halo around your foreground. That might have been the issue with that clip of the guy in front of the building.

[Stevet]

It looks great Jim... Hey, did you buy an EX1?

[Jim Arthurs]

Hi Steve, yes I did buy one! I couldn't be more thrilled with the picture quality, and the whole keying quality came out of left field... I wasn't expecting much and was pleasantly surprised.

Here's a layered photoshop file with uncompressed and 35mb codec layers if folks want to mess with their own keys...

http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_35mb_vrs_8bit_uncompressed.psd

And here's a layered file with a center crop from a 1080p frame showing the difference between the default detail in PP1 vrs detail off. Simply apply your own sharpening techniques to see the that you're not giving up anything by leaving it off.

http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_DETAIL_ON&OFF.psd

[Sumfun]

I was going to start a new thread, but saw this one... here you go... comparing the native codec vrs. the HD-SDI out...



Thanks for posting the test, Jim. Both keys look good, but I think the right side definitely looks better, especially through the hair and fingers. (Although there were not much hair and fingers on the right side, so it was hard to compare). Just curious, which keyer did you use?

Also, this is the first time that I've heard about turning off the detail. Could you please explain how it makes your picture worse?

[Jim Arthurs]

This was done in After Effects with Keylight. I generally do these sort of tests over a medium grey as it will point out all flaws.

I think the EX1's codec is amazing for the file size, as that uncompressed stream it's compared to is many, many, many times larger in bandwidth for a fairly marginal return in quality.

Most of the posted EX1 clips you'll find around have detail on, and you can detect it as an artificial band of dark between areas of contrast. It's something that is very much a visual cue of "bad video".

A few posts down I have an A/B compare with detail on/off... take a look at that and you can see that the edges have this issue when detail is on.

With the EX1, the detail feature doesn't ADD real detail to the image, it just does an artifical sharpening effect to the image.

This is a bell that can't be un-rung when shooting and doing key work, so turn it off. Add sharpening later after the matte is pulled if that's your preference.

In all honesty, I've never shot one frame of footage with detail "on" with this camera, other than that chart test to prove to myself that real resolution isn't impacted by turning it off.

[Nicky]

Nice one Jim, great input.:thumbsup:

I thought you were an intense VFX guy, its interesting to see you dont mind all this compression and rolling shutter stuff. To be honest I also dont mind it too much... what can I say - Im a sucker for bright, rezzy, high dynamic range clean images!

I think its quite easy to ignore the disadvantages of CMOS when the images are so beautiful and cheap.

[Christopher Barry]

Jim, totally awesome information and presentation. Thank you.

[rubento]

Jim,

Could you share your PP settings for your Chroma key test? I found it very, very good...

Thank you

[Sumfun]

Thanks for the explanation, Jim. After enlarging the picture, I can definitely see the black (and white) lines added by the detail feature. Also, I was impressed by how the chroma blur cleaned up the colors at the edges.

[Lenilenapi]

Jim,

When I played the video clip I saw this kind of magenta or purple fringing around his hair. What's that about?
It was unacceptable to me. Is that in your clip or something that was introduced in getting through compression or whatever to the web?

- Lenny Levy

[Jim Arthurs]

Hi Lenny... I didn't work this shot more than ten minutes or so, and with the only goal to demonstrate the relative value of the codec vrs. uncompressed... Some of that is un-tweaked spill issues when using blue, but that's not the whole story.

To better put this in context, the lighting was a couple of shop fixtures with 3000K CRI 73 bathroom tubes. Not a full spectrum light source, and certainly responsible for some of the problem.

I will say this, from actual experience... I've had much more difficulty with the RED under perfect 3200K lighting using bluescreen due to noise, than with the EX1 even in this slipshot setup.

While we're on this topic, what "video camera" would do as well as this in anything remotely close to the cost of the EX1 even hampered with its native codec? :)

FYI, I had a go with the HPX2000 last week and it was a complete let down for green screen work when recording to DVCPRO 100 1080 30p.

[Lenilenapi]

Jim,
If you ever finish tweaking that shot or have another similar shot with similar hair issues, please post that.
My situation is that i don't know a damn thing about cutting screens. I just need to be able to convince my clients that this camera can do it. If I showed them that shot, they would ask about spill.

I would love to be able to just say - here look at this !

Thanks as always for your generosity with information.

By the way what are you using for PP set-ups? I've given up on the cine gammas and the auto knee and am trying to just come up with a nice knee that starts around 93 or so and std Gamma 2 or 3.

- Lenny

[Jim Arthurs]

Lenny, I've reuploaded a more polished version as the same file, feel free to check it out. There's a tiny hint of magenta, but I wanted to stay in the realm of a one layer same setting key for the whole shot. FYI, the key settings are exactly the same on both sides, 35 Mb and uncompressed. The only difference is a slight gamma inbalance between the HD-SDI capture and the codec mapping.

My overall impressions are that the codec blooms the blue channel a bit compared to the HD-SDI, and that the HD-SDI retains a bit more resolution for detail in the chroma areas and there is less noise in the marginal areas of transparency. As I'd hope for that increase in data rate.

However, if this is acceptable to you or not, it does nothing to help you or your client out in the decision making process. All it does is say that some guy in Colorado shooting under the worst possible lighting conditions did an okay matte on a solid grey background. The only thing that would matter to your client is how good you light, and your people comp. I would put forth that I don't think the camera will be the thing standing in anyone's way.

Finally, since this is very important to you, and you're not yet seeing something worth selling in the area of keys... what camera ARE you currently using that is adequate for your client? I can dig you up some F900, some RED and Viper frames that suck toe compared to this latest version of the comp...

As to PP settings, I'm experimenting now with some, and will be in the snow tommorw for some tests... I don't know enough about the Cine settings to like or dislike them yet.

Regards

Jim

[Jim Arthurs]

Jim,

Could you share your PP settings for your Chroma key test? I found it very, very good...

Thank you

Gosh, it was literally in the first hour of testing the camera, the only modification was turning off the detailing entirely and -3db. Currently I've shot a couple jobs with a PP modified to -4 black and Gamma set to STD 3 and that seems a fine "safe" set of options. -4 on the black seems to push the darkest black (lens cap closed) down to -just- before crushing at 0 IRE... -5 might be too much.

As I mentioned in the last post, I'm going to experiment with some Cine settings tomorrow... I'm just using this as a starting point for the tests...

http://provideocoalition.com/index.php/awilt/story/review_sony_pmw_ex1_1_2_3_cmos_hd_camcorder/P3/

[Lenilenapi]

Jim,

Forgive me for being so dumb about keying.
I take it you are saying that with a little bit more work and an extra layer you could easily remove the remaining magenta fringing right. That's typical for a shot with fine hair anyway isn't it. I take it those are never easy.

- Lenny

[Jim Arthurs]

Yes with another shot at it, and that color (magenta) is an issue only with blue screen, not green screen. It is NOT a camera issue that could be fixed by better glass, etc. I'm sure the low CRI lights with their incomplete spectrum are bumming up the mix more than anything else.

The only real point I'm making with this shot is that if you factor all the handicaps into the equation; low color temp forcing the camera to drive the blue more than normal, incomplete color spectrum, blue screen being more difficult than green because of blue channel noise in general... it just seems to me that we've got something a cut above anything in the price range and a bit higher.

Or as I like to say to clients on set when I pull a quick key to test the lighting setup, "This is as bad as it will get. It will only get better.".

[basspig]

Jim, you did that on blue, not green screen? Egads! Sure beats the heck out of HDV! You're inspiring me to play with the green screen keying again.

[HVXguy]

Egads! Better than the RED??
Do you have your RED now??
Could you do a side by side with the Sony & the RED?

[booth]

Egads!

(just wanted to join in)

[Jim Arthurs]

The problem with testing against the RED is that it is constantly a moving target with the firmware updates. When I did a day with RED 232 doing blue and green screen tests, it was clear that the camera was very challenged in the blue channel under tungsten illumination. This impacts not only blue but green screen work as well. Here is a test I did that shows clearly those issues...

http://ftp.datausa.com/imageshoppe/outgoing/RED_232/blue_channel_compare.png

Now, as I understand, there's new improvements in the firmware and also different quality settings for REDCODE, so it's all up in the air again and would require re-testing. Which is a great idea when I get a free day to do so.

At the time, there was no question, even with the super-sampling down from 4K Bayer to 2K 4:4:4 that the EX1 4:2:2 is/was a competitive performer in the keying department under tungsten lighting.

But hey, feel free to have a go with our RED key source footage (girl with bunny)... just download the clips, fire up REDCINE or REDALERT and process your collective hearts out...

http://www.redrelay.net/owners/0232/

[Stevet]

Jim, it appears from your results the RED ONE camera's CMOS sensors seem to be balanced for 5600K.

I imagine the EX1 sensors are balanced for 3200K.
I believe this is one reason why the EX1 would fair better in the blue channel.
Those RED ONE results surprised me. Thank you for posting.

If so, I wonder why they chose a 5600K balance?
3200K lighting is plentyful, not to mention all the conventionals in standard lighting.