Interesting new light field lens is coming

[dabateman]

This may not work for UV at all. Maybe not even IR.

But the idea behind it is quite fascinating.  Also with good Zeiss optics and mirrors,  might be worth the kickstarter price. 

https://www.k-lens-one.com/

 

You will need a 135 format sensor.  I like that it will be F-mount and adapted to all cameras.  They might even just make a bunch with specific mounts, without needing adapters. 

 

The resolution might be an issue though. 

My Df with 16Mpixel goes down to 1.55Mpixels. Then up to 3Mpixel with multiple image stacking and maximum 6Mpixels with their AI algorithm,  which I haven't seen and might not believe. 

A 24Mpixel camera goes to 2.7Mpixels.

A 42Mpixel A7R2 goes to 4.7Mpixels.

A 45Mpixel goes to 5Mpixels, which might be the sweet spot. 

The Sigma fpL or Sony A7R4 sensor goes to 7Mpixels.

That could be an ideal combination,  the small Sigma fpL mounted without any source of vibrations. 

 

Have to wait and see more images. But if I want it, would definitely need to mount it on a higher resolution sensor than my only 135 format Nikon Df.

[OlDoinyo]

My German is rusty enough that I had difficulty reading the website in its entirety. But I gather that there is an intermediate focal plane inside the lens with some sort of reprojection behind (I am unsure whether passive or active.) With all that glass thickness, I would not hold much hope for UV usability. IR might be a different matter, depending on what internal filtration might be present. Interesting, though. I wonder what the intended use case really is.

[Kai]

Interesting.
3-D from different perspectives at the same time. I've already tried stereo photography with two images. That is also impressive (although a different goal is pursued here, namely 2-D images with z-axis information per pixel).
Actually, this should also work with a focus stack and Helicon focus. I just shot a small scene with two lenses with an 85 at f / 1.8 (4k video). Helicon makes such a 3-D model out of it. Unfortunately, my manual focus rotation was too inhomogeneous. The result looked unusable. Maybe it's better with in-camera focus stacking?

 

What would be really exciting:

Take each of the 9 images through a different narrow band filter from UV to NIR and thus record a hyperspectral image!

A lens with an integrated filter set for less than 2000 euros - it no longer seems out of reach! 

[Andy Perrin]

1 hour ago, OlDoinyo said:

My German is rusty enough that I had difficulty reading the website in its entirety. 

You can go to the menu and select English as the language, OlDoinyo! 

 

Here:

https://www.k-lens-one.com/en/home

--

 

David, the price is just $2,059 with 50% discount! You will just need to sell your child, which I'm sure will thrill your wife after the purchase of that quartz lens awhile ago...but who needs a family anyway when we can have cool toys?

[Andy Perrin]

1 hour ago, Kai said:

Interesting.
3-D from different perspectives at the same time. I've already tried stereo photography with two images. That is also impressive (although a different goal is pursued here, namely 2-D images with z-axis information per pixel).
Actually, this should also work with a focus stack and Helicon focus. I just shot a small scene with two lenses with an 85 at f / 1.8 (4k video). Helicon makes such a 3-D model out of it. Unfortunately, my manual focus rotation was too inhomogeneous. The result looked unusable. Maybe it's better with in-camera focus stacking?

 

On the 3D issue: if you want to make a 3D image from multiple 2D views, this is easy to do with photogrammetry software. I have Agisoft Metashape software, and I've been using it happily for months to make 3D images of gravestones for purposes of reading faded inscriptions.

 

Example: 

 

I took 16 photos from different angles. This is just a sample of one of them.

 

Metashape aligns the points in all images and calculates depth by triangulation to make a "sparse point cloud". You prune out bad points using their tools to estimate point quality. Bad points are gray here and are filtered out. I also masked the background to avoid spurious matches and capturing leaves and other things not of interest.

 

Then you make a dense point cloud from the sparse cloud and mesh the model. Here is the fully meshed model.

"HERE LYES BURIED

the BODY OF Mr.

IOHN PARLEY

WHO DIED MAY Ye 2nd

1725 IN the 56

YEAR OF HIS AGE

-----------------

IF YOU WIL LOOK IT MAY APPEAR

HE WAS THE FORST BURIED HERE."

 

(And indeed, his was the oldest grave in the cemetery, because he gave the money for the cemetery...)

 

 

Once you have the dense point cloud, it's also possible to output depth maps for each of the individual images. For example, the original image above has depth map like this:

 

Anyway this is much much cheaper than a $2000 lens! The tradeoff is that you can't do any video and you need multiple images. It would work fine with IR and UV, and indeed the professional version of Metashape includes direct support for multispectral imaging even.

 

 

[dabateman]

2 hours ago, Andy Perrin said:

 

 

David, the price is just $2,059 with 50% discount! You will just need to sell your child, which I'm sure will thrill your wife after the purchase of that quartz lens awhile ago...but who needs a family anyway when we can have cool toys?

Ya most likely would not go over well. But their design does really interest me. I wonder if it could be simplified to a filter added to any lens. That might work,  but would be tricky.  Maybe just a filter to add to one specific focal length. Still tricky,  kaleidoscope filters are just garbage.  But it might be possible with small tinny lens elements fused together. 

The rotating multiple spectral idea would also be fun. 

I see this lens having a better use case for video,  as currently that's really hard to duplicate cheaper.  I am just not big into video. 

Also we will have to see were this goes. With computational photography and a cheaper manufacturing method,  this might come down to the masses.

[Kai]

On 11/21/2021 at 7:13 PM, Andy Perrin said:

 

On the 3D issue: if you want to make a 3D image from multiple 2D views, this is easy to do with photogrammetry software. I have Agisoft Metashape software, and I've been using it happily for months to make 3D images of gravestones for purposes of reading faded inscriptions.

 

Example: 

 

I took 16 photos from different angles. This is just a sample of one of them.

 

 

Metashape aligns the points in all images and calculates depth by triangulation to make a "sparse point cloud". You prune out bad points using their tools to estimate point quality. Bad points are gray here and are filtered out. I also masked the background to avoid spurious matches and capturing leaves and other things not of interest.

 

 

Then you make a dense point cloud from the sparse cloud and mesh the model. Here is the fully meshed model.

"HERE LYES BURIED

the BODY OF Mr.

IOHN PARLEY

WHO DIED MAY Ye 2nd

1725 IN the 56

YEAR OF HIS AGE

-----------------

IF YOU WIL LOOK IT MAY APPEAR

HE WAS THE FORST BURIED HERE."

 

(And indeed, his was the oldest grave in the cemetery, because he gave the money for the cemetery...)

 

 

 

Once you have the dense point cloud, it's also possible to output depth maps for each of the individual images. For example, the original image above has depth map like this:

 

 

Anyway this is much much cheaper than a $2000 lens! The tradeoff is that you can't do any video and you need multiple images. It would work fine with IR and UV, and indeed the professional version of Metashape includes direct support for multispectral imaging even.

 

 

Fascinating! Long, dark winter evenings are coming soon ... :)
Many thanks, Andy!
 

[dabateman]

On 11/21/2021 at 1:13 PM, Andy Perrin said:

 

On the 3D issue: if you want to make a 3D image from multiple 2D views, this is easy to do with photogrammetry software. I have Agisoft Metashape software, and I've been using it happily for months to make 3D images of gravestones for purposes of reading faded inscriptions.

 

Example: 

 

I took 16 photos from different angles. This is just a sample of one of them.

 

Metashape aligns the points in all images and calculates depth by triangulation to make a "sparse point cloud". You prune out bad points using their tools to estimate point quality. Bad points are gray here and are filtered out. I also masked the background to avoid spurious matches and capturing leaves and other things not of interest.

 

Then you make a dense point cloud from the sparse cloud and mesh the model. Here is the fully meshed model.

"HERE LYES BURIED

the BODY OF Mr.

IOHN PARLEY

WHO DIED MAY Ye 2nd

1725 IN the 56

YEAR OF HIS AGE

-----------------

IF YOU WIL LOOK IT MAY APPEAR

HE WAS THE FORST BURIED HERE."

 

(And indeed, his was the oldest grave in the cemetery, because he gave the money for the cemetery...)

 

 

Once you have the dense point cloud, it's also possible to output depth maps for each of the individual images. For example, the original image above has depth map like this:

 

Anyway this is much much cheaper than a $2000 lens! The tradeoff is that you can't do any video and you need multiple images. It would work fine with IR and UV, and indeed the professional version of Metashape includes direct support for multispectral imaging even.

 

 

Andy how many images does this software need to get a 3D model? Can this software be used to change the focus point or expand the depth of field? I see its $180, slightly cheaper than $2260 for this light field lens. 

 

I quickly made a four element fused silica 18mm focal length lens with the four elements able to cover my m43rds sensor. As they are just 6mm in diameter.  A single 200mm fused silica biconvex lens seems to help sharpness in front of it without impacting the field of view. I still need to test a concave element in front of it.

I will see if I can snap some images in the various UV wavelengths and create a new thread. 

 

[Andy Perrin]

Dabateman, you need 3 points of view (photos) to uniquely specify a location in space. So that is the absolute minimum. But both common sense and math suggest the more the better because it increases precision. My basic rule of thumb is about 12 pics per gravestone, and I only bother with the front. 
 

The Metashape software is intended primarily for making 3D models so the depth maps are a consequence of the 3D modeling rather than the main purpose of the program. It does let you save them (for any of the photos). To do things like refocus the photo would need additional software, but a depth map is a depth map no matter how it is obtained. Just import into the same software they will use with the lens above, or any other refocus software (I still have Lytro Desktop which would let me do that probably, although I haven’t tested it). I see someone has figured out a photoshop hack with the Lens Blur filter to make it use a user-supplied depth map, so if you have a PS subscription, you are set. 

 

I will say that depth map accuracy increases if you take photos with a 1/3-of-a-screen camera shift between each shot. One thing I’m curious about for that lens is how much depth accuracy you can get with all 9 pics taken from almost the same angle. 

[dabateman]

Well after quite a lot of digging.  It looks like there is free software that can generate depth maps and even machine learning is stepping up to the plate to do it with single images.

But I could use DMAG (many version based on different methods,  not updates) as in DMAG4 is for a single image, whereas DMAG5 uses stereo pair.

Also stereo photo maker looks to also create depth maps using a modified DMAG5/9 algorithms. 

Then I need to investigate the software I actually own. It looks like Topaz lens effects not only can give you a fake blur image. But also load a depth map to more precisely pick out an image plane.

Some discussion also looks like Affinity photo might also be capable.  

Or use GIMP, were much of this started.

It always comes down to what do you actually want to do and what look are you after.  

This looked to be a hot field back in 2013 to 2015. But some of the fire has cooled down.  I guess the demise of 3D TV and the wigglegram crazy has subsided. 

[Andy Perrin]

David, what are you trying to actually do? It’s true you can make a fake depth map from a single image but that’s not a thing that will give any accurate distances. Photogrammetry is the gold standard on that regard, micron-level resolution can be achieved. However the fake depth maps are probably still usable if you just want to refocus images. As I said, the photoshop lens blur filter will take a depth map in the alpha channel that can be supplied by the user. I’m sure Topaz would also be fine in that regard. 
 

I guess for me the depth map itself is the point, since my purpose is reading inscriptions, and I need accuracy. The refocusing of images gets old fast. I tried it when the Lytro came out — I have the gen 1 Lytro camera. But it actually is fun with only a very restrictive subset of photos. 

[dabateman]

33 minutes ago, Andy Perrin said:

David, what are you trying to actually do? 

Yes exactly.  The more I think about it the more I realize this is nothing that I want to do.

But I did learn something.  Adding a bicx element infront of my dual or now quad element assembly for 3D stereo images in UV improved the image quality. I was pushing the single elements forward or back, but adding an other larger element that covers them all is better.  It helps to stare at others lens assemblies. 

[Andy Perrin]

Cool!

[Alaun]

I tried to understand, what they do (but I have only two mirrors):

 

Instead of looking at the real object, they use a virtual image to look at.