What I love about gaussian splats is the way they degrade - instead of a hard cutoff or LoD changing spheres into cubes etc., they get increasingly "dreamy" - the basic idea is still there, just less detailed.
I know feelings about AI are mixed. But when AI can dream up gaussian splats in real time, from a prompt, and do refinement as you get closer to things... That's going to be pretty bonkers.
That's kinda what NERFs are (neural radience fields). They actually preceeded this Gaussian story, with Gaussians coming in and outperforming them. Maybe they'll merge later for something even better, I don't know enough about them.
Yes they are image generators. You want image generator generators.
A diffusion style process generating gausians instead of pixels. You could possibly do nerfs that way, but it would be effectively generating a trained network. If you managed to do that it would have broad application throughout the field of AI.
I could see a kind of fun game / design tool / worldbuilding where you get a blurry world and you describe what you are seeing, and it comes into focus. The game world, mechanics, aesthetic, and playstyle build as you form your view. A sort of fog of war meets rorschach game.
This will be the future of a class of 3d Game. the prompt may not be text however.
An input of a kind of schematic representation of what the designer wants would be better. It may resemble a storyboard or a collection of organised notes that large projects tend to already use.
Fully generative could probably do some cool things, but people will still want to bring their peronal vision to life.
Curious, why wouldn't the future be a full world model like Google's Genie? It just renders every pixel so someone could still make their vision come to life via a prompt too.
Yeah, when you describe that, I picture Wave Function Collapse to generate a map schematic... And then a text prompt, and some style photos the designers want it to match.
I don't know anything about them but it's a cool effect. At least on this strawberry, you're not zooming in but rather traveling closer. I don't see the increasing (made up) detail you'd expect from a zoom, we sort of pop through the skin into an invented interior.
its funny that sky features like clouds and blue patches end up being fit very close to ground level because there isn't a difference in perspective to cue in the algorithm that the skybox should be tall, I wonder if there is any way to incorporate the fitting algorithm with simultaneous lidar data about true distances of things to allow scenes to be viewed from further up
This is like a beautiful little miniature. It's cool to see gaussian splatting applied to a detailed small thing versus a big scene.
I have a question about perhaps the most boring aspect of this strawberry: the license. you write, "You can download it under CC BY license, but attribution is appreciated rather than required." IANAL, but I'm pretty sure you can't license your work CC-BY and then waive the BY requirement in the description. Rather, you'd have to license it with something more permissive like CC0 and request attribution if you want attribution to be optional. Is that right?
That new Laowa 180mm macro lens is amazing. So sharp, free of chromatic aberrations, compact, and I like the long working distance and reduced perspective distortion that the 180mm focal length offers.
I usually use PostShot.. but the quality was not very good. I want to try LichtFeld but my Graphics Card has too little memory.. so I reached out on twitter, some people ran tests and Mykhailo got some better quality out of it so I took his training. You can d/l the COLMAP dataset for free and try yourself.
This one keeps crashing the browser after it reaches 100%. Safari/iOS, iPhone 13. I was able to navigate and use a few of the other ones linked to from the comments, though. Curious.
(I'm ... still not sure what I'm looking at on TFA, and whether or not my browser configuration fails to fully present the site as intended.... OK, if all you're seeing is a blurred image of a strawberry, yes, you'll want to enable a bunch of JS resources. I'm using uMatrix, several hosts must be enabled.)
Ah the timeless joy of falling through the floor geometry.
Seriously though - it's breathtaking.
The first guy who figures out the bridge between splats and dynamism - animation, editing, responsiveness - is going to be one of the immortals of 3d design.
There are many ways to represent 3D data, but animations really only work properly with polygon meshes (e.g. triangle surface meshes or volumetric tetrahedral meshes).
But the mesh is itself an abstraction, you just need to build that bridge.
We've been leaning away from pure polygons for decades, anyway. Vertex skinning, SDFs, volumetrics, simulation, and a lot more.
The meshes in a From Software game are for exmple hilariously simple, most of the animation is force simulation to make the famous "frizzles" that they like.
Apples model to generate Gaussian splats from a single image. Takes about 30 seconds on an M1 Pro.
It falls apart once you move too much, but for a little side-wiggling or a second-eye view for VR, it's great. And looks a lot better than the old approach of depth map + vertex shaders that I use in https://github.com/combatwombat/tiefling. But ml-sharp has 2.6 GB weights, a bit too big to run in the browser.
As I have learned about Gaussian Splatting just a few weeks ago, I have (perhaps funny/naive/stupid) question: is there any progress or at least theoretical chance to have dynamic lighting?
There are some works on doing this directly e.g. https://arxiv.org/abs/2601.23065 but getting accurate materials is a challenge for anything more than diffuse.
AI-based relighting will no doubt start working soon.
out of the box, I imagine that surfaces can be lit, but probably not shadowed correctly. (structures aren’t solids, more like particles in 3d space)
it could look like the real-time lighting of an old game engine on rather modern assets. (quake 2-3 era)
or perhaps some "occlusion pre-pass" could be done to create a voxelized sparse volume from the splats that set a "voxel opacity value" for each to absorb light? (not far from how prebaked GI works nowadays)
note: not an expert on rendering, just a nutjob that did stuff in opengl in the old days.
Lots of translucent blobs composited to produce the appearance of a strawberry.
There is no mesh or model. The visual surface of the strawberry could be made up of blobs spaced far apart physically and not where the surface appears to be.
This is why they are called radiance fields, they model the light not the geometry.
Practically the blobs positions/rotations can be constrained to better physically match the geometry of a strawberry.
I'm not sure i agree. The blobs are exactly where the surface appear to be because they are constrained by multiple viewing angles.
Otherwise the splat would fall apart as soon as the viewing angle is changed slightly (Which it absolutely does in many examples on supersplat, you cannot really create an out of distribution view with 3GS, it's not magic)
Yes, my statement was loose. The blob doesn’t really have a position since it is theoretically an infinite distribution in 3 space.
It has a mean, and that mean doesn’t have to lie on the surface, consider the case where the mean is deep inside the strawberry but its spike contributes to the surface appearance (e.g a seed could be represented this way, or it could be represented by a small well-oriented blob on the surface, the optimiser doesn’t care)
Others have provided details about how it works. I suggest zooming way in on that image and you'll start 'breaking though' the surface and that'll help you get an idea of how it works. Important thing is there is no defined geometric surface ("mesh"). Also important to know is that it's very, very hard to get a good splat without taking a ton of photos at different angles. It's also really, really easy to create a crappy looking splat. But when it's done right, it's a marvel
When you say its very very hard to create a good splat, what do you mean? And what is good? I would say that strawberry is very detailed and its a good splat. I also kind of like the way some of the 'rougher' splats look. I feel like they'd work well in a car racing simulator.
Oh, that would be amazing! The photos are also available and some people definitely take it for a test ride. However.. This dataset with the stacking is far from generic.
The filesize of a 3d animated splat is seemingly very small, and the method enables ~arbitrary FPS. But it seems the setup required to record it is still huge and expensive, which limits its usefulness.
Even with that there are some interesting use cases, eg. I'd love to be able to watch concerts this way, and freely move around the stage and crowd from any angle.
There is a faint sensation of translucency, I wonder if that's an artefact of the process, or if it's the actual optics of the surface layer if the strawberry...
It's an artefact unfortunately. Gaussian splats have no concept of refraction and have a hard time dealing with reflection. Highlights, usually so so. Something I always wrangle.
I wonder if that is somehow related to how you seem to have very strong studio light for the subjects? Somehow the bright light penetrating deeper into the material or something like that
The light doesn't need to be completely flat like what you'd do when scanning it more traditionally, for meshing. The problem is when the highlight is sharp and reflects it's surrounding. You could remove it, with cross polarization.. But then it wouldn't look realistic anymore.
Well, at certain angles you actually get very noticeable gaps in the strawberry that are visible when rotating it, but almost invisible when static. I think it's mostly due to that.
Does anyone know if it's already possible to build gaussian splat of the person that moves/rotates from the single camera? (I.e. to use sequence of frames to reconstruct occluded parts of the body for other frames)
As I understand it, Gussian Splats aren’t a method of constructing missing data in the sense you’re asking here.
But there are other, well established, methods of generating missing data in convincing manner, that are beyond my field of expertise or interest to be able to repeat here.
With the added benefit that if I’m wrong, being wrong can be a great way to spur discussion 8)
I tried making one, but I couldn't make through the camera position tracking bit, software was super unintuitive. Very interested in gaming applications for this tech, but still waiting for it to be more approachable from a layman's point of view.
One thing you can do with this that works quite well is use it to "decrop" (widen the viewing angle) of a video. This is very useful for stabilization which usually involves cropping. Here's an example: https://x.com/i/status/2051504427287404568
Once they get quicker to train, I expect this to be a popular use of them
My intuition is that in theory focus stacking should not be necessary as preprocessing step for 3dgs (or photogrammetry). Does anyone know if there is any recent developments in this regard?
Focus stacking generally is not perfect process and can lead to artifacts/errors and I'd imagine those can then compound when stacked images are used for 3dgs. Also the image focus actually provides some depth data in itself that could be useful?
My take.. at a macro scale, the dof is usually so small, that it's hard to get a reliably track. So you'd need some sort of way to tell that these stacked photos belong into a series, and then you sort of are doing focus stacking :-)
I do think the alignment algorithm could be improved. Maybe the approaches I linked could be used to make a much more robust focus stacking algorithm, that also corrects for 3D geometry. That would be really cool!
If you don't focus stack and try to train on partially unfocused images, the optimizer will try to match the rendered view to be also partially unfocused.
You would have to mask out the blurry areas for each image. I guess one could just implement a feature where the optimizer only optimizes gaussians within the sharp distances relative to the camera.
Other way of looking at the question is if you could make focus stacking better by using the full multi-view dataset? Afaik focus stacking essentially does depth estimation so it seems like multiple views would help with that.
Another way would be some kind of 4d GS where one dimension is the focus distance. But I'd guess the renders would inherently have shallow depth then, which is less useful usually.
I'm wondering if the splat community has decided this paper is valuable -- https://github.com/fraunhoferhhi/Self-Organizing-Gaussians -- looking at all the detail in the strawberry splat made me wonder how small one can get the download, and what the current state of the art is for compression.
Looking at all the outdoor scenes that you can walk around, I wonder how long until we start seeing this in places like Google Maps/Earth, as a replacement for the low-res 3D renderings we have now.
I guess the number of samples required to generate a GS is the constraint now, but maybe that will get solved.
rendering them is also non trivial from what i understand, since the millions of indivudal point positions need to be recalculated all the time while moving
This was also my first thought when I zoomed in into the strawberry. I wonder if you can achieve a microscope like effect with a more suitable setup. E.g. better lighting, zoom, lens, etc.
I have done 2x macro (an ant).. and want to try 5x.. but as you get closer, the depth of field becomes really shallow. You can do focus stacking but you risk that the individual areas in focus are less ideal aligned and the tracking can't make any sense out of the geometry anymore.
The flesh of the strawberry is very slightly translucent. By taking images from many angles, you can derive some information about the interior. This is not a separate step, it is simply the optimizer finding splats that contribute the right amount of color and light at various viewing angles. Of course those splats aren’t really going to look right to anyone who has ever eaten a strawberry, because they are only derived from the input images and not your past memories of a real strawberry.
You’ll also see the same type of problems in any incomplete part of any scene. Go poke around any of them and you’ll find places where there wasn’t enough information to make a satisfying reconstruction. For example, the back of the moon <https://superspl.at/scene/2ac8f423>, the fridge in the coffee shop <https://superspl.at/scene/6a0c3ccf>, underneath the footbridge in the forest <https://superspl.at/scene/23ebe85c> (or indeed anything in the distance), etc, etc.
From the link: "Shot from 90 perspectives, 88 focus stacked images each. Nikon Z8, full frame, f/7.1, exposure 1/160, ISO 100, Laowa 180mm macro lens, with LED light and bluescreen."
Insane!
i made some decent splats (admittedly much lower quality than this) by taking video with my iphone13 mini and then chopping it up into individual frames via ffmpeg
There is actually still a Lytro type camera for industrial application available: https://raytrix.de/
Too low res, too expensive for my use case unfortunately.
I wonder would a good, sharp needle and thread make for good mounting for a soft object like this? Thread the needle, pass it right through the strawberry and the secure the thread on something above and below. As long as the strawberry doesn't slide down the thread (hopefully a strawberry is light enough friction would hold it in place!)
I glue mounted the strawberry on three nails and used pins to secure it.
I have to think about your idea.. I don't think friction would be enough to hold it in place.. it would probably be hard to knot the top onto something too, as that's where the light is: https://i.imgur.com/vIjw6pc.jpeg
But I'm always experimenting with the mounting, thanks for the inputs.
Assuming that the person that did this has not tried that. If you look at the setup photos, the grape is resting on a couple of nails. This suggests that many different things have been tried.
I want some math Phds to sit down with the Corridor Crew guys and figure out how to make Gaussian Splat reconstruction way better, the way they came up with CorridorKey.
The splat of the studio has the perfect amount of detail. It looks like you're streaming from the camera direct to the computer usually? How do you check the progress / quality while capturing? Seeing the (great) results makes me more curious about the process of creating now.
Thanks! I use the computer only to setup the shot, adjust lighting make test stacks of different angles. If everything is good, I disconnect and capture to the memory stick. During the shooting I only adjust the beginning or end of the rail a bit, so everything is in focus at some point. Triggering (manually) the rotary disk still happens from the computer.
Imagine if we start designing GPUs around this technology as opposed to vectors. Imagine what voxel engines would look like. Would love a simulated experience or a small scale that theorizes about this.
A Youtuber that records jookin (a mid-South form of dance) uses Gaussian splats for his videos to bring the dancers out into the forefront and hold center stage, it's a neat trick.
Beautiful.
What I love about gaussian splats is the way they degrade - instead of a hard cutoff or LoD changing spheres into cubes etc., they get increasingly "dreamy" - the basic idea is still there, just less detailed.
Take for example this scene:
https://superspl.at/scene/e721ea7c
If you navigate closer to the trees, things around you become blurry - as if the very fabric of reality unraveled.
I know feelings about AI are mixed. But when AI can dream up gaussian splats in real time, from a prompt, and do refinement as you get closer to things... That's going to be pretty bonkers.
That's kinda what NERFs are (neural radience fields). They actually preceeded this Gaussian story, with Gaussians coming in and outperforming them. Maybe they'll merge later for something even better, I don't know enough about them.
Sure, but NERFs were trying to match your input photos and poses, not some arbitrary prompt, if I understand correctly.
Yes they are image generators. You want image generator generators.
A diffusion style process generating gausians instead of pixels. You could possibly do nerfs that way, but it would be effectively generating a trained network. If you managed to do that it would have broad application throughout the field of AI.
What would this look like in practice? A net that outputs weights for a new net to use?
Couldn’t you “uncurry” such a process to have only a single network?
NERFs have significantly higher image quality than 3D Gaussian Splatting or more recent similar techniques, though they are much slower to render.
This one month old video did a reasonable job of getting my entirely ignorant self relatively up to day on NERFs and Gaussian Splats:
https://youtu.be/X8yRlA7jqEQ
I could see a kind of fun game / design tool / worldbuilding where you get a blurry world and you describe what you are seeing, and it comes into focus. The game world, mechanics, aesthetic, and playstyle build as you form your view. A sort of fog of war meets rorschach game.
This will be the future of a class of 3d Game. the prompt may not be text however.
An input of a kind of schematic representation of what the designer wants would be better. It may resemble a storyboard or a collection of organised notes that large projects tend to already use.
Fully generative could probably do some cool things, but people will still want to bring their peronal vision to life.
Curious, why wouldn't the future be a full world model like Google's Genie? It just renders every pixel so someone could still make their vision come to life via a prompt too.
Yeah, when you describe that, I picture Wave Function Collapse to generate a map schematic... And then a text prompt, and some style photos the designers want it to match.
This sounds like it could be a great concept for a future sequel to LSD: Dream Emulator
If I'm not mistaken that is the inspiration for one of Alt-J albums
They remind me a lot of the 'memories' scene in Minority Report:
https://www.youtube.com/watch?v=arTIRgdEb1g
I don't know anything about them but it's a cool effect. At least on this strawberry, you're not zooming in but rather traveling closer. I don't see the increasing (made up) detail you'd expect from a zoom, we sort of pop through the skin into an invented interior.
its funny that sky features like clouds and blue patches end up being fit very close to ground level because there isn't a difference in perspective to cue in the algorithm that the skybox should be tall, I wonder if there is any way to incorporate the fitting algorithm with simultaneous lidar data about true distances of things to allow scenes to be viewed from further up
That's what Google Street View should be like more.
I like that they are sort of between a photo and a 3D model. Nothing quite like it.
Is there a specific name for this “dreaminess” effect?
Splattenuation
The Setup: https://i.imgur.com/o0hgybh.jpeg https://i.imgur.com/mcNiomp.jpeg https://i.imgur.com/vIjw6pc.jpeg https://i.imgur.com/nzOwmSC.jpeg
This is like a beautiful little miniature. It's cool to see gaussian splatting applied to a detailed small thing versus a big scene.
I have a question about perhaps the most boring aspect of this strawberry: the license. you write, "You can download it under CC BY license, but attribution is appreciated rather than required." IANAL, but I'm pretty sure you can't license your work CC-BY and then waive the BY requirement in the description. Rather, you'd have to license it with something more permissive like CC0 and request attribution if you want attribution to be optional. Is that right?
I'm pretty sure it's allowed, just as saying "do what you want with it" is an informal license, it's just a little pointless.
That's what I thought too.. supersplat has no way to set it to CC0 otherwise.
That new Laowa 180mm macro lens is amazing. So sharp, free of chromatic aberrations, compact, and I like the long working distance and reduced perspective distortion that the 180mm focal length offers.
Can you elaborate why you chose slang-splat over let's say Lichtfeld studio? How does it compare to the other splat-training tools?
I usually use PostShot.. but the quality was not very good. I want to try LichtFeld but my Graphics Card has too little memory.. so I reached out on twitter, some people ran tests and Mykhailo got some better quality out of it so I took his training. You can d/l the COLMAP dataset for free and try yourself.
Just wow!
As I scrolled through the website, I was even more impressed with this one though!
https://superspl.at/scene/c67edb74
This one keeps crashing the browser after it reaches 100%. Safari/iOS, iPhone 13. I was able to navigate and use a few of the other ones linked to from the comments, though. Curious.
I just wanted to do a quick size comparison...
If I'm reading Chrome right, that's 171 MB, for the website, and the data.
If I'm doing the math right, that's 40 seconds worth of the bandwidth from Netflix, at its highest rate.
Beautiful and quite neat to 'walk' up into the ceiling and look down. Feels like you're either quasimodo or God, depending on how nuts you are.
shame the bishop behind looks a bit flat though ;-)
I built PlayCanvas in 2011 to power video games. Here we are in 2026 and it's powering strawberries.
Thank you for your service!
This is one of the most endearing things about open source IMHO, the way people can find novel uses for it.
I completely agree with you. I'm constantly bowled over by the creativity of the SuperSplat/PlayCanvas communities.
For those unfamiliar with Gaussian Splatting: <https://en.wikipedia.org/wiki/Gaussian_splatting>.
(I'm ... still not sure what I'm looking at on TFA, and whether or not my browser configuration fails to fully present the site as intended.... OK, if all you're seeing is a blurred image of a strawberry, yes, you'll want to enable a bunch of JS resources. I'm using uMatrix, several hosts must be enabled.)
Wow this is a time killer... ended up here: https://superspl.at/scene/ff1d0393 beautiful!
It’s amazing this runs perfectly smooth on my iPhone 12 mini
That is indeed a very cool scene being about to wander around and still have decent resolution
Ah the timeless joy of falling through the floor geometry.
Seriously though - it's breathtaking.
The first guy who figures out the bridge between splats and dynamism - animation, editing, responsiveness - is going to be one of the immortals of 3d design.
There are many ways to represent 3D data, but animations really only work properly with polygon meshes (e.g. triangle surface meshes or volumetric tetrahedral meshes).
But the mesh is itself an abstraction, you just need to build that bridge.
We've been leaning away from pure polygons for decades, anyway. Vertex skinning, SDFs, volumetrics, simulation, and a lot more.
The meshes in a From Software game are for exmple hilariously simple, most of the animation is force simulation to make the famous "frizzles" that they like.
You pretty much just need a representation that can be constructed reasonably and interpolated.
Is it me, or does the backside of that berry not look too good?
Also interesting: https://github.com/apple/ml-sharp
Apples model to generate Gaussian splats from a single image. Takes about 30 seconds on an M1 Pro.
It falls apart once you move too much, but for a little side-wiggling or a second-eye view for VR, it's great. And looks a lot better than the old approach of depth map + vertex shaders that I use in https://github.com/combatwombat/tiefling. But ml-sharp has 2.6 GB weights, a bit too big to run in the browser.
Any idea what kind of vram is needed to run this on Nvidia?
[delayed]
It runs natively on the Photos app in Vision Pro, which shares 16GB of ram with GPU and CPU.
The Python process peaks at about 15 GB, so maybe a 24 GB GPU works. But others seem to have problems wih Nvidia: https://github.com/apple/ml-sharp/issues/84
As I have learned about Gaussian Splatting just a few weeks ago, I have (perhaps funny/naive/stupid) question: is there any progress or at least theoretical chance to have dynamic lighting?
There are some works on doing this directly e.g. https://arxiv.org/abs/2601.23065 but getting accurate materials is a challenge for anything more than diffuse.
AI-based relighting will no doubt start working soon.
out of the box, I imagine that surfaces can be lit, but probably not shadowed correctly. (structures aren’t solids, more like particles in 3d space)
it could look like the real-time lighting of an old game engine on rather modern assets. (quake 2-3 era)
or perhaps some "occlusion pre-pass" could be done to create a voxelized sparse volume from the splats that set a "voxel opacity value" for each to absorb light? (not far from how prebaked GI works nowadays)
note: not an expert on rendering, just a nutjob that did stuff in opengl in the old days.
There can be some issues with shadowing yeah, especially if you render with splatting/rasterization, but its fine if you raytrace I think.
I read [1], but I still don't quite know what I'm looking at. My guess is a 3D model reconstructed from lots of detailed pictures?
[1] https://en.wikipedia.org/wiki/Gaussian_splatting
Lots of translucent blobs composited to produce the appearance of a strawberry.
There is no mesh or model. The visual surface of the strawberry could be made up of blobs spaced far apart physically and not where the surface appears to be.
This is why they are called radiance fields, they model the light not the geometry.
Practically the blobs positions/rotations can be constrained to better physically match the geometry of a strawberry.
I'm not sure i agree. The blobs are exactly where the surface appear to be because they are constrained by multiple viewing angles.
Otherwise the splat would fall apart as soon as the viewing angle is changed slightly (Which it absolutely does in many examples on supersplat, you cannot really create an out of distribution view with 3GS, it's not magic)
Yes, my statement was loose. The blob doesn’t really have a position since it is theoretically an infinite distribution in 3 space.
It has a mean, and that mean doesn’t have to lie on the surface, consider the case where the mean is deep inside the strawberry but its spike contributes to the surface appearance (e.g a seed could be represented this way, or it could be represented by a small well-oriented blob on the surface, the optimiser doesn’t care)
This video explains how Gaussian splatting works: https://www.youtube.com/watch?v=X8yRlA7jqEQ
This was great, thanks for sharing
Others have provided details about how it works. I suggest zooming way in on that image and you'll start 'breaking though' the surface and that'll help you get an idea of how it works. Important thing is there is no defined geometric surface ("mesh"). Also important to know is that it's very, very hard to get a good splat without taking a ton of photos at different angles. It's also really, really easy to create a crappy looking splat. But when it's done right, it's a marvel
When you say its very very hard to create a good splat, what do you mean? And what is good? I would say that strawberry is very detailed and its a good splat. I also kind of like the way some of the 'rougher' splats look. I feel like they'd work well in a car racing simulator.
https://youtube.com/watch?v=X8yRlA7jqEQ is how I learned about them. They're really cool!
Here’s a good 2 minute explainer https://youtu.be/HVv_IQKlafQ
Noting the permissive license, could we be witnessing the birth of a new Utah teapot or Sponza atrium?
Oh, that would be amazing! The photos are also available and some people definitely take it for a test ride. However.. This dataset with the stacking is far from generic.
I'm really interested to see what folks can do with animated Gaussian splats: https://youtu.be/X8yRlA7jqEQ?si=dXeHa03jO7MTBNLA
The filesize of a 3d animated splat is seemingly very small, and the method enables ~arbitrary FPS. But it seems the setup required to record it is still huge and expensive, which limits its usefulness.
Even with that there are some interesting use cases, eg. I'd love to be able to watch concerts this way, and freely move around the stage and crowd from any angle.
Great video. I was about to share it here myself.
There is a faint sensation of translucency, I wonder if that's an artefact of the process, or if it's the actual optics of the surface layer if the strawberry...
It's an artefact unfortunately. Gaussian splats have no concept of refraction and have a hard time dealing with reflection. Highlights, usually so so. Something I always wrangle.
I wonder if that is somehow related to how you seem to have very strong studio light for the subjects? Somehow the bright light penetrating deeper into the material or something like that
The light doesn't need to be completely flat like what you'd do when scanning it more traditionally, for meshing. The problem is when the highlight is sharp and reflects it's surrounding. You could remove it, with cross polarization.. But then it wouldn't look realistic anymore.
Well, at certain angles you actually get very noticeable gaps in the strawberry that are visible when rotating it, but almost invisible when static. I think it's mostly due to that.
Does anyone know if it's already possible to build gaussian splat of the person that moves/rotates from the single camera? (I.e. to use sequence of frames to reconstruct occluded parts of the body for other frames)
I watched this 1 month old video two days ago:
https://youtu.be/X8yRlA7jqEQ
It kinda splatted my mind.
As I understand it, Gussian Splats aren’t a method of constructing missing data in the sense you’re asking here.
But there are other, well established, methods of generating missing data in convincing manner, that are beyond my field of expertise or interest to be able to repeat here.
With the added benefit that if I’m wrong, being wrong can be a great way to spur discussion 8)
I tried making one, but I couldn't make through the camera position tracking bit, software was super unintuitive. Very interested in gaming applications for this tech, but still waiting for it to be more approachable from a layman's point of view.
One thing you can do with this that works quite well is use it to "decrop" (widen the viewing angle) of a video. This is very useful for stabilization which usually involves cropping. Here's an example: https://x.com/i/status/2051504427287404568
Once they get quicker to train, I expect this to be a popular use of them
My intuition is that in theory focus stacking should not be necessary as preprocessing step for 3dgs (or photogrammetry). Does anyone know if there is any recent developments in this regard?
Focus stacking generally is not perfect process and can lead to artifacts/errors and I'd imagine those can then compound when stacked images are used for 3dgs. Also the image focus actually provides some depth data in itself that could be useful?
This is an interesting question. There has been some attempt to model the camera better than just a pinhole camera, that could work. https://dof-gaussian.github.io/ https://github.com/leoShen917/DoF-Gaussian
My take.. at a macro scale, the dof is usually so small, that it's hard to get a reliably track. So you'd need some sort of way to tell that these stacked photos belong into a series, and then you sort of are doing focus stacking :-) I do think the alignment algorithm could be improved. Maybe the approaches I linked could be used to make a much more robust focus stacking algorithm, that also corrects for 3D geometry. That would be really cool!
If you don't focus stack and try to train on partially unfocused images, the optimizer will try to match the rendered view to be also partially unfocused.
You would have to mask out the blurry areas for each image. I guess one could just implement a feature where the optimizer only optimizes gaussians within the sharp distances relative to the camera.
Other way of looking at the question is if you could make focus stacking better by using the full multi-view dataset? Afaik focus stacking essentially does depth estimation so it seems like multiple views would help with that.
Another way would be some kind of 4d GS where one dimension is the focus distance. But I'd guess the renders would inherently have shallow depth then, which is less useful usually.
> Another way would be some kind of 4d GS where one dimension is the focus distance.
That seems like the way to do it, someone has to be working on that.
searching...
https://dof-gaussian.github.io/
Dany, this is so cool.
I'm wondering if the splat community has decided this paper is valuable -- https://github.com/fraunhoferhhi/Self-Organizing-Gaussians -- looking at all the detail in the strawberry splat made me wonder how small one can get the download, and what the current state of the art is for compression.
Thanks! We have two compressed formats, the sog by PlayCanvas and spz with sparkjs. Both now support LODs and compress really well.
Looking at all the outdoor scenes that you can walk around, I wonder how long until we start seeing this in places like Google Maps/Earth, as a replacement for the low-res 3D renderings we have now.
I guess the number of samples required to generate a GS is the constraint now, but maybe that will get solved.
Google already has some splats on google maps,
https://youtu.be/gXug7Kb3p4I
rendering them is also non trivial from what i understand, since the millions of indivudal point positions need to be recalculated all the time while moving
Turns out, GPUs are amazing at this.
Someone: Please combine microscopy with gaussian splatting.
Don't know if this would be in your wheelhouse, but for very nice macro splats, check out the work by Dany Bittel: https://danybittel.ch/macro.html
For example this bumblebee: https://superspl.at/scene/cf6ac78e
Edit: I completely missed that this was posted by him (:
Incredible that specular highlight isotropy vectors are somehow encoded.
This was also my first thought when I zoomed in into the strawberry. I wonder if you can achieve a microscope like effect with a more suitable setup. E.g. better lighting, zoom, lens, etc.
I have done 2x macro (an ant).. and want to try 5x.. but as you get closer, the depth of field becomes really shallow. You can do focus stacking but you risk that the individual areas in focus are less ideal aligned and the tracking can't make any sense out of the geometry anymore.
Word of warning to mobile users, this website completely crashed my Firefox instance on Android.
How did they decide what the inside of the strawberry looks like? Because that seems very incorrect.
The flesh of the strawberry is very slightly translucent. By taking images from many angles, you can derive some information about the interior. This is not a separate step, it is simply the optimizer finding splats that contribute the right amount of color and light at various viewing angles. Of course those splats aren’t really going to look right to anyone who has ever eaten a strawberry, because they are only derived from the input images and not your past memories of a real strawberry.
You’ll also see the same type of problems in any incomplete part of any scene. Go poke around any of them and you’ll find places where there wasn’t enough information to make a satisfying reconstruction. For example, the back of the moon <https://superspl.at/scene/2ac8f423>, the fridge in the coffee shop <https://superspl.at/scene/6a0c3ccf>, underneath the footbridge in the forest <https://superspl.at/scene/23ebe85c> (or indeed anything in the distance), etc, etc.
Great work! There are more awesome splats on the author's profile page: https://superspl.at/user?id=danylyon
From the link: "Shot from 90 perspectives, 88 focus stacked images each. Nikon Z8, full frame, f/7.1, exposure 1/160, ISO 100, Laowa 180mm macro lens, with LED light and bluescreen." Insane!
And it only takes 20 Minutes to shoot all 7920 photos, the Z8 is crazy fast.
i made some decent splats (admittedly much lower quality than this) by taking video with my iphone13 mini and then chopping it up into individual frames via ffmpeg
Where's the Lytro camera when you need one?
I was really pulling for OTOY to keep making light field capture and display technology...
There is actually still a Lytro type camera for industrial application available: https://raytrix.de/ Too low res, too expensive for my use case unfortunately.
You might want to throw that one away :)
It is in fact still on it's mount and slowly rotting / molding.. for a second capture :-)
Yup. It's rotten on the other side. Or maybe lead poisoned
Sorry if I fell for Poe's law, but just for clarity, the strawberry's rotten underside is most likely missing splats in the rendering.
Of course it is. That's where the joke comes from...
Edit. TIL Poe's law
not sure if that's a joke but i think that's just the effect of the strawberry being placed on a glass/plastic surface to be filmed from underneath
What happened to the bottom of that poor strawberry?
It was mounted at the bottom.. and I can't quite reach it with my camera. Might have to try some two pass way.
I wonder would a good, sharp needle and thread make for good mounting for a soft object like this? Thread the needle, pass it right through the strawberry and the secure the thread on something above and below. As long as the strawberry doesn't slide down the thread (hopefully a strawberry is light enough friction would hold it in place!)
Anyway, very cool splat, fair play
I glue mounted the strawberry on three nails and used pins to secure it. I have to think about your idea.. I don't think friction would be enough to hold it in place.. it would probably be hard to knot the top onto something too, as that's where the light is: https://i.imgur.com/vIjw6pc.jpeg
But I'm always experimenting with the mounting, thanks for the inputs.
Mount it on a needle/skewer, it should let you capture it in one pass.
Assuming that the person that did this has not tried that. If you look at the setup photos, the grape is resting on a couple of nails. This suggests that many different things have been tried.
How about green screen + rotate the strawberry on a skewer
Gaussian splat casualty. The bottom looks like partially missing from the reconstruction.
When you cut the splat in half, result is either fuzzy fog or sort-of fibrous crystals. As depicted here.
I want some math Phds to sit down with the Corridor Crew guys and figure out how to make Gaussian Splat reconstruction way better, the way they came up with CorridorKey.
Gaussian splats look really good from a distance, but as soon as you zoom in, they really fall off a cliff :/
Can you show the setup?
(Can we do a Gaussian Splat of the setup of the photograph for the Gaussian Splat of the Strawberry?)
I just posted some images.. I have actually done my studio, (quality is.. let's say painterly ;-): https://superspl.at/scene/0a3916cd
The splat of the studio has the perfect amount of detail. It looks like you're streaming from the camera direct to the computer usually? How do you check the progress / quality while capturing? Seeing the (great) results makes me more curious about the process of creating now.
Thanks! I use the computer only to setup the shot, adjust lighting make test stacks of different angles. If everything is good, I disconnect and capture to the memory stick. During the shooting I only adjust the beginning or end of the rail a bit, so everything is in focus at some point. Triggering (manually) the rotary disk still happens from the computer.
THAT IS AMAZING!! Seriously, impressive work! Thanks for the link.
How can I start creating Guassian Splat of my environment with my 360 camera? Is it possible to do that with local software?
This doesn't work at all for me (Linux desktop, tried with Firefox and Chrome). I only see "fullscreen-extended blurry thumbnails" of the splats.
Works for me with Firefox 140.10.2esr on Devuan 5.
Works for me on Debian 13, Firefox 140.10.2esr-1~deb13u and Strix Halo gfx
If you can, please check any output/errors on the developer console?
Makes me wonder if we'll ever see a video game where the world is created by set designers irl.
Sim racing is at least kind of like that.
Very nice. did you segment the images before?
Yes it's shoot in front of a bluescreen.
this is awesome, I wonder what's under there, looks black, maybe thats where they mounted and rotated the strawberry...
Yes mounting.. and I can't quite reach all the way from below.
Imagine if we start designing GPUs around this technology as opposed to vectors. Imagine what voxel engines would look like. Would love a simulated experience or a small scale that theorizes about this.
Lovely! How was the mechanical setup to ensure that all those shots are consistent, and how long did it take?
I posted some pictures.. takes only 20 min!
Impressive, but my poor GPU is melting :)
What? KIRI Engine makes splats. I always wondered what 3DGS might mean.
Yes. I knoweth what "splats" are: They are splats of fuzzy blobs on the display surface.
Really terrible performance in Chrome, I'm getting between 10-15fps.
What's the matter?
It appears to be rendering locally, my fans started spinning up when I was wandering through another image on the site.
Where else would it render??
I mean streaming renders are a thing but they're very much the exception not the rule
ew the bottom is moldy
A Youtuber that records jookin (a mid-South form of dance) uses Gaussian splats for his videos to bring the dancers out into the forefront and hold center stage, it's a neat trick.
https://www.youtube.com/watch?v=k0MGoFaIiWs - WARNING - heavily adult lyrics and a loud starting "MOTHERFU*AAAAAAS" right off the top.
That doesn’t really look like Gaussian splats to me- but this video, Helicopters by A$AP Rocky, uses the technique well:
https://youtu.be/g1-46Nu3HxQ?si=y_u257DzgOg2QV83
What makes you say this is using Gaussian splats? It looks like simple motion-based cropping to me.