LART results thread

[Steve Gibby]

This thread will be for members of the LART team to post images and links to motion sequences from LART that have been processed and cleared for posting.

On this thread shortly:

- Gavin Greenwalt and Jeff Kilgroe will be posting some images here shortly of greenscreen work. Nice stuff!

- Mike from VUZE is preparing some h264 sequences to post on the new GibCor channel of VUZE. Links will be posted here.

- Some cool r3d files from various scenes we shot will be be posted for download.

We have a vast amount of r3d files, pics, frames, and test result summaries to sort through, categorize, and post. It won't happen all at once, but we'll peck away at it and progressively post more and more results. :weight_lift: :help:

Enjoy!

 

[Ken Corben]

Fellow redusers,

Please allow this thread for LART team members to post results ONLY. The idea is to keep this thread simple and easy to read and find links. Please post comments. requests and other items on a new thread under the LART section of reduser so that everyone can benefit from an easy to follow thread that links to all the information, download links, pics, etc.

Thanks,

Sharky

 

[Gavin Greenwalt]

Ok I'll be posting a 16bit Tif in a few minutes here so I'll start off with a bit of commentary on the r3d -> Tiff process itself.

There was some talk a few days ago about whether DPX, Tif, OpenEXR or TGAs provided the best keying results and I think I understand why the OP thought that Log DPXs keyed the best. Using default settings they do because REDCine and (I'm guessing REDAlert) hardly touch the color settings when exporting a Log DPX. Now that's not to say that Tifs and OpenEXRs won't provide quality images as well, it just means you need to do some of the normal prep grading that you would do in the compositor inside of REDCine before export.

I've noticed that the Red channel tends to have less dynamic range after correcting for white balance. This makes it pretty clippy and punchy and REDCine's exposure controls have a habit of completely clamping dark values for whatever reason so even a little bit of white balancing can cut out up to 2 stops of DR from the RED Channel.

The provided Tif is almost exactly what I used for my 'demo key'. Even though the bluescreen was lit at 3.5k tungsten I set the WB to 5k and then individually tweaked each channel until I felt like I wasn't clipping or crushing any channels individually. Funny... after balancing it looks a LOT like a viper's natural color balance... :whistling:

Edit:
Bluescreen/Lace/BlondeHair/DarkHair/GreenShirt/Shiny Jewlery shot:
MediaFire:http://www.mediafire.com/?atmumcsgdxc
REDRelay: http://www.redrelay.net/lart/im.thatoneguy/Flirty_Lace_ChromaPlate_4k_04088.tif

Rough example comp and bluescreen JPG in next post...

[Gibcore approved]

 

[Gavin Greenwalt]

Ok here is a before/after of a comp I did this morning in between jobs along with some BTS photos of the setup in case you want some reference.

Bluescreen was lit with 2 Disistis (Sp?) and 2 kinos for patching any holes. Screen itself was some sort of felt material clamped to speed rail. Pretty low tech but it got the job done for the most part.

Obligatory disclaimers: This is just a single key with a quick spill supress to give you guys some ideas and test out various experiments so that I can figure out what's actually going on. More about getting from start to finish to see what challenges and road blocks there are than producing a pretty image at the end, although the processes are quite similar. With more time you could get a lot more out of it by breaking it down and individually extracting all the bits you want. It's still pretty choppy, spill in hair needs more individual attention.. Grain was just for fun... am not responsible if this blows up your computer etc... blah blah blah. NOT ANYWHERE NEAR FINAL

IMO the blue channel is really the REDCode's achilles heal and if you do end up needing to shoot on a blue screen I would look at using the other channels whenever possible as luma or diff keys to extract the fine details. I've run this shot through about 4 different keyers this morning (the one I posted was just a lone primmatte with supression added later.) All of the straight chroma keyers returned pretty smudgy keys because of the wavelet, but the more difference based keyers such as IBK did a bit better.

 

[ChrisLyon]

Remember that footage hosting is available at RedRelay without having to use regular public hosts.

Public upload is via FTP:

address: perennialmedia.com
username: lart@perennialmedia.com
password: reduser

The files will be located at http://www.redrelay.net/lart

Also, be sure to give information about your shots like lens, ISO/ASA, Framerate, etc.

 

[Gavin Greenwalt]

Here is the same model on green and blue for comparison. Again exposed/white balanced for maximum range.

Link:
MediaFire:http://www.mediafire.com/?em0hkm3tnnv
RedRelay:http://www.redrelay.net/lart/im.thatoneguy/SplitChroma_ChromaPlate_1k_01099_v01.tif

---

We have more bluescreen plates and tests yet to come this is only the beginning.

Edit: Whoops I dropped the ball on taking camera notes for the green screen tests. Focal length is going to be variable since it was shot on a zoom. I metered it before hand at T5.6 @ 400ISO but I don't know what the operator ended up shooting with for those two shots.

Edit2: had a few minutes in between tasks to work on a key. ((-1)Garbage mask + Difference + IBK).

Edit3: It occured to me that most of you probably don't have a clue what "IBK" means. IBK stands for image based keyer. I would say it's a keyer developed by digital domain however to call it a keyer while accurate is a bit misleading. It's really just a standard 20 year old difference keyer (compare these two images and tell me in a single value the difference between each part of them) combined with some handy tools to make it easier to work with. It's an impressive piece of "low-tech reimagined" which produces under the right circumstances quite astonishing results comparable or surpassing the latest fancy algorithms.