KOMODO 4-CAMERA STITCH ARRAY...help!!!

[H Dewey]

Original thought was to shoot a Hydrogen 2D/4-6 unit array. Now thinking of switching to 4 cam Komodo array. I'd be a bumbling idiot if I undertook this without getting help from anyone who has experience or ideas in this sort of thing.

Anyone, whether experienced in stitching stills or video... save me from becoming a bumbling, blithering idiot. '-)

 

[Gregory Karydis]

Phil,
This one's for you!

 

[Björn Benckert]

I would make a metal cheese plate where the cameras ar bolted down as close as possible to each other in a cross configuration. And make sure to use. conehead screws or somthing similar to make sure the cameras mount identical every time you build up the rig. That is esential then if doing it right you can use the same stich setup every time.

Then for stitching ptgui is simply the king.

Stitch a set of stills from the array and export uvmaps to use in any compositor. like nuke, fusion, flame or what ever.

Pt gui is extremly good at eliminating lens oddities etc. And you can also render ou blend mattes etc.

Kommodo is by no comparison a better tool for 360 than any other of the red cameras as its so small but most of all, it got global shutter, that is essential to handle motion in a good way.

Im not sure the can sync lock but I would be surprised if they cant do it trough the comport.

 

[Phil Holland]

Phil,
This one's for you!

Already designed my rig. That's so 2019 Glass was the harder problem to solve. But there's good options.

 

[Björn Benckert]

Already designed my rig. That's so 2019 Glass was the harder problem to solve. But there's good options.

If cost is no issue, then I would say for the given sensor size best picture option for 4 camera rig if there is no demand for vertical 360 but only horizontal I would go for 4 x Ultraprime 8mm lenses.

But then yes, pl is possibly not the best option nice possibly to go for a EF lenses so focus and iris can be controlled trough the camera dont know what the best ultra wide EF is.


We did a horizontally 360 project back in EPIC MX days for bank of america that came out stunningly good in terms of picture quality using the UPs. But as 4 of those is quite big and costly.

A low cost option for 360 is the sunnex lenses. For those you also only need 3 cameras to do a full 360 dome vert and horizontally and then you got 1/3 image as overlap. As the sunnex shoots 182deg, 3 instances more than cover. Sunex has no iris, but with behind lens filtering that should not be much of an issue. they do em in ef an nikon mount. We remounted one into PL which is quite easily done.

http://www.superfisheye.com

 

[Phil Holland]

You couldn't pay me to use that 8mm in 2020. Way better solutions in the last decade.

I prefer Entaniya Fisheyes lately for that sort of thing, more expensive, but notably higher quality than any Sunex lens I have here.

Here's a grab from a 16K sequence in the mid-2000s:

^ that one is about a 100mm wide image plane, slow crawl dolly in over the river via crane.

And here's one that didn't make the cut from the 12K Monstro Array (we didn't end up using deep night footage for the pieces):

^ that is about a 60mm wide image plane, though a bit motion blurry because we were buzzing the bridge with the heli.

 

[Björn Benckert]

Whats wrong with the UP 8mm?

I know the sunex is not the sharpest of glas, but actually holds up quite well especially considering the price and how little of it that comes to use if stitching 3 or more. But its not utilizing the whole sensor.

And just as the sunex, is the Entaniya not a bit overly wide if using a 4 camera rig, no?

8mm UP to me looks smashing on komodo sized sensor and if shooting it in 4 directions you get something like a 25% overlap if I remember correctly so final resolution from 4x6k becomes something like a 15k image if not downscaling / only doing the destortion. The Entaniya gets a little less than half that, no?

 

[Gregory Karydis]

Thanks Phil!
The only input I have on the subject, is that I'd rather go for something rectilinear, much like the RF native Laowa 15mm f2 Zero-D https://www.venuslens.net/product/laowa-15mm-f2/
Or, if you want to break the bank, the Canon EF 11-24mm f/4L

The Laowa should net you a 84° HFOV at the FF equivalent of 20mm
While the Canon should become a 101.7° HFOV on the wide end at the FF equivalent of 15-32mm

Haven't tried the Laowa, but the 11-24mm on my 5DSr is perfect!
I'll try and do some still stitching to get you more info on this.

 

[Phil Holland]

Whats wrong with the UP 8mm?

There's nothing inherently wrong with any lens. If you like it, you like it. Love me some funky glass actually.

In the case of the 8R it's a bit heavy, on older design in the face of more modern corrected optics, and longer from the image plane than necessary compared to some smaller higher quality lenses that can do the equivalent.

I only highlight this as we are discussing a bit more of a technically challenging way of filming and little things play larger roles sometimes. Even the way bokeh draws can effect stitching if you plan on including that concept in your shot.


I mostly use rectilinear lenses for larger arrays with the intention of a planar projection. Though I've used fisheyes too, but those are mainly used for my HDR, 360, IBL rigs. 360x180 equirectangular stuff.

Oddly most of goals with arrays have been more to explore higher resolutions and larger formats to receive the benefits from all of that, though a lot of the early stuff was VFX and exhibition driven. Now I'm mostly deploying it to make a ridiculously high quality 8K via downsample or 16K image. About once or twice a year a 32K job shows up which usually is a bunch of displays or panel based system. That's rarer and harder work honestly if you're moving the many cameras through space and time. Once you get beyond 15 camers it's a big, big process. Array work already is demanding to say the least. I know a couple jobs that sat in post for well over a year painting madness out.

Oddly with my VFX background it sort of highlights really testing things out and getting it right on set rather fixing it in post, even though something like this is going to post no matter what. Nail the rig. Test it thoroughly and you'll get there. Unstructured arrays are indeed harder work if you're dealing with a lot of moving and parallax. But when everything is in line you're treated with really nice high quality images.

 

[H Dewey]

Thanks Phil, Björn and Greg. You've given me much to think about... especially since I'll be trying to line up a square image with 4 units. I've suspected the lenses (matched) would be the hard part and to Phil's point, wanted to combine all images to gain a higher resolution.

And speaking of resolution, do I need to work with a high resolution display and high power computer system (to pixel peep when aligning the four images away from the edges?) I assume proxies will not give one the ability to precisely join the four images?

Phil: Oddly with my VFX background it sort of highlights really testing things out and getting it right on set rather fixing it in post, even though something like this is going to post no matter what. Nail the rig. Test it thoroughly and you'll get there. Unstructured arrays are indeed harder work if you're dealing with a lot of moving and parallax. But when everything is in line you're treated with really nice high quality images.

Me: Words to live by... and have nightmares about getting it right. '-)

 

[Phil Holland]

Hardware will really depend on your workflow and needs, but you will 100% need to do some pixel peeping. My stitching workflow is 16-bit uncompressed, so big data and data rates. That's not for everybody though for sure.

Quality control is key for pulling of a seamless stitched sequence. Most people who even saw my NY stuff in 2017 didn't know it was stiched, even though that actually was only about 80% done. There are stitching errors in there actually, but YouTube compression and downsampling hid my evils and I was determined to release that pre-NAB that year, I was the single source for post on that, but the final piece was all tidy about a month later. As well as the crazy birds flying through frame shot I published later. That one wasn't fun. Didn't some weird adaptive transform mesh junk, comp a smidge, water is always "fun", but a pixel accurate stitch makes pains go away. The hard stuff is moving close and far away from objects and moving through space as well. Further away = much easier stitch. Stationary camera = much easier stitch.

Beyond that, patience. You'll have some long renders if you're trying to get up there. For the 32K stuff we deliver tiled deliverables, which isn't fun. 16K I can delivery in a single file and there is stuff that can play that back. Most of it is split into 4K or 8K tiles though if it is being displayed at higher resolution.

 

[H Dewey]

Hardware will really depend on your workflow and needs, but you will 100% need to do some pixel peeping. My stitching workflow is 16-bit uncompressed, so big data and data rates. That's not for everybody though for sure.

Quality control is key for pulling of a seamless stitched sequence. Most people who even saw my NY stuff in 2017 didn't know it was stiched, even though that actually was only about 80% done. There are stitching errors in there actually, but YouTube compression and downsampling hid my evils and I was determined to release that pre-NAB that year, I was the single source for post on that, but the final piece was all tidy about a month later. As well as the crazy birds flying through frame shot I published later. That one wasn't fun. Didn't some weird adaptive transform mesh junk, comp a smidge, water is always "fun", but a pixel accurate stitch makes pains go away. The hard stuff is moving close and far away from objects and moving through space as well. Further away = much easier stitch. Stationary camera = much easier stitch.

Beyond that, patience. You'll have some long renders if you're trying to get up there. For the 32K stuff we deliver tiled deliverables, which isn't fun. 16K I can delivery in a single file and there is stuff that can play that back. Most of it is split into 4K or 8K tiles though if it is being displayed at higher resolution.

I've been under the assumption I'll have to shoot as stationary (heh, probably just set all the lenses to f/11 and go from there. '-)

I'll just move the array around, closer, farther to get my variety of shots... plus, I'll likely trim the stitched image to use parts of it to cut down to a smaller final image size for exhibition... also, the script I have in mind doesn't have a lot of action shots so hopefully that translates to easier post.

But I gotta admit... hardware for workflow scares me.

 

[Mark L. Pederson]

It is close to impossible to beat MISTIKA SGO tools for optical flow stitching in addition to the some of the best lens and geometry correction tools you can find . And you can create a template for your custom rig and dramatically accelerate stitching process. Monthly subscription model allows you to pay-per-project. https://www.sgo.es/mistika-boutique/

 

[H Dewey]

It is close to impossible to beat MISTIKA SGO tools for optical flow stitching in addition to the some of the best lens and geometry correction tools you can find . And you can create a template for your custom rig and dramatically accelerate stitching process. Monthly subscription model allows you to pay-per-project. https://www.sgo.es/mistika-boutique/

A cursory look at the website suggests they are deeply involved in making stitching easier. I like the idea of the one month subscription at reasonable rates. Thanks Mark.