I'm not gonna lie, but our development team is in the middle of trying to 3d model items and models, but because none of us can 3d Model(mainly because we're only 14😅) We in a bit of a hole. Becuase a good multi-player horror game needs good quality, and YOU my friend made my day, Thank you our team wishes you the best of luck for the future 🤩💪
The biggest feature that would help photogrammetry workflows would be separating the speculative map from the basic material. or even creating materials from gaussian Spalatting. The way I see it, a Gaussian Spalt model is a “3D latent space” and a KI algorithm (probably even a cheap comvolutional approach) could explore to find out where primary reflection “move” over the surface and where ambient color stay stable even if perspective changes. The classic approaches to capture reflective maps in photogrammetry like photometric stereo or polarized flashes are very time consuming. It would be awsome if ai could “learn” where an object is “reflective” “translucent” or “refractive” by walking the 3d space of a Gaussian splat capture 🙏
Thank you for such a detailed comment, just for this comment, let me share some exciting news we are going to reveal soon :) So, basically, we are working on a noval inverse-rendering technique where you can not only 3D scan completely transparent and reflective objects and get mesh, if there is stuff inside the transparent object, for example, an ice cube inside a transparent bottle, this new technique can also reconstruct both the bottle and the ice cube inside, it's cool and I can't wait to show it to the public in the coming month
@@KIRI_Engine_App that sounds really exciting! I wonder if the turntable platform setup with fixed camera will work well for this. It is necessary for Gaussian splatting to move the camera relative to the object AND light sources? Or will the inverse rendering technique will cover stable camera, rotating object and stable lights (in relation to the camera, not the object) ?
Great job by the way just needed to say that. As you likely know though the reasons 3d artists need a mesh and why mesh export is important the ability to at some point relight, change materials and fully integrate GS into a shot comprising of both kinds of assets in a believable way. Great first steps
That's a great comment, thank you Ethelbert! Yes, personally speaking (it's Jack here BTW), I'd be happy to see if we can have 3DGS achieve 70% of what photogrammetry can do by the end of next year, and for that goal, we are working to solve relighting, shadowing casting, physic etc problems on 3DGS
Oh wow. This is it! Now we can finally say adieu to photogrammetry . 3D modelling will become more fun.. Exporting as obj is cherry on top🙂. Thank you so much guys.🙇♂
Thank you so much man :) But I'm afraid that most people are gonna have to stay with photogrammetry for a little longer :P Because in this mesh-inclusive 3DGS version, the mesh quality is still not quite there comparing with photogrammetry, but we'll get there in the future
Yes, we checked out SuGaR the first day it came, but it was way too bad to be used in an actual product. This algorithm is a heavily engineered version of NeuS, but it does the job much better :)
oh that is really gaming changing . i didn't talk enough Kiri engine in my yt channel , will do . thanks for your work and i love your videos as your app
This is a great start, and I commend your efforts on outputing something useable. I think you know these objs are going to be dense meshes of hot garbage thought, and almost useless for any 3d modeling purposes (or worse than useless). The actual novel work begins if y'all choose to tackle the non-trivial challenge to output useful geometry, I know it might be too much, but good luck if y'all take this on, it would be a world first basically (Instant Meshes is like a 20% solution) (of course 'useless' = recreating by hand would be faster than cleaning it up)
Hey there! That's a very insightful comment. I agree that even with mesh from 3DGS, it won't be as usable as other methods, but we just need to keep pushing the technology and adaption because if we don't, who will... If you are looking for more usable models, our photogrammetry and NeuS models are good for you, we have some serious stuff like quad mesh retopology and PBR generator to improve the usability, feel free to give it a try :D
There is two use cases (or technologies) in 3D reconstruction where Gaussian Splatters probably cannot compete with a mesh as 3d representation format but at the same time could trigger massive change in how nowadays they are used as in their visual quality (textures). They are Lidar mapping (large scale mapping) and 3D scanners. It is very difficult a Gaussian Splatters or it mesh can compete in precision and quality with such devices, used i.e. in survey or reverse engineering. In such cases the mesh you are able to get, can be used to aling the gaussian splatter representation with the lidar/3Dscanner mesh, and reproject the GS texture in their mesh.
import mesh into 3D software - create a texture that pipes color/texure from g.splatting renderer (capture an envelope of splats from around that mesh zone) - use the 3D mesh as a mask in 3D app.
@@LeeBrenton Oh wow, thank you so much for such a technical comment. We know how to write code, but we are not really 3D artist, so every time we see this kind of comment it made us understand a bit more about the realm. Please feel free to join us on Discord here, we'd love to talk with you more to see if we can further help with the pipeline: discord.gg/Jt2JHG7BAc
@@KIRI_Engine_App i will buy kiri engine pro to support you on this path. one thing I'm wondering is if there is a difference in the quality of the pro plan. and besides, you usually make videos on gaming platforms, kiri engini promises a very important future for e-commerce, I think it would be useful if you make videos about it.
For use cases where accurate measurements of digitalized objects are needed, do you see the future in: A.) More and more methods emerging to convert the gausian splats into very accurate point clouds and mieshes (regarding geometrical and spatial fidelity), or B.) more and more methods emerging to use the splats without conversion? Do you see any chances that Gaussian splats will replace traditional photogrammetric models, for example for surface roughness estimation, object and 3D feature (like cracks in buildings or rocks) detection?
Great question GamingShiiep! If we are prioritizing accurate measurements, then mesh will always be the king, but if we are prioritizing visualization, then I do believe 3D Gaussian has a chance. And I don't think 3D Gaussian Splatting will ever replace photogrammetry, but it will find its own place where photogrammetry is limited, for example, it's impossible to photogrammetry hair or thin leaves, but 3DGS can do it, so I think in the future creators get to use different representation within the same project to make the 3D scene more realistic
@@KIRI_Engine_App Thanks for your answer! I have a couple more questions (sorry!), but they might give you some ideas: 1) Assuming we use the same input frames/photos, hyperparameters and framework for the point cloud generation using photogrammetry and 3DGS, will we end up with the exact same positions of the points? 2.) Is the Mesh creation in Kiri "simply" a linear interpolation between the points or are other methods used to create more points by something like intercepts of various degrees of the Splats? Since things are going super fast and I'm lacking the math skills behind programming and enhancing such a solution (for now), I'm sure I don't need to keep such an idea for myself. 3.) When I assessed the quality of my photogrammetric model using video frames vs photos, the quality difference was huge and much better with much less photos. I strongly assume that it's because of noise, compression, and color depth etc. (I used my Google Pixel 7 pro cameras at 60 fps and 4k). Why does everybody in the context of NeRF and 3DGS always talk about using video then? For photogrammetry a higher number of video frames did not make up for the lack of quality compared to fewer photos (600 frames vs 60 photos. HUGE difference). Is that different for 3DGS or for NeRFs? If so, why? 4.) Why are most of these 3DGS and NeRF show case videos always using such wild camera movements and not just slow and smooth "camera rides"? I have absolutely no Idea why that would be. Also, isn't everything at the end rendered for interactive viewing on a screen? Isn't a photogrammetry model that you look at (either mesh or point cloud) also rendered to be seen on the screen?
@@GamingShiiep Wow, these are very well-thought-out questions! Let's do it: 1. Yes, we will end up with (almost) the same positions of the points if we use the same frameworks and parameters as inputs. The reason I wouldn't say the exact same is that even if you use the same input for photogrammetry two times, you won't get the exact same points. This is just how random the algorithm is 2. Yes and no, basically there are two steps of point cloud generation in our photogrammetry. There's an initial point cloud generation we call camera pos calculation using SFM techniques. After the first step, we will end up with sparse point cloud; Then we use the sparse point to extract more points from the photos, this step is called dense point cloud generation. After that, we use some mathematical appoint to align the points to make sure all point alignments are as smooth out as possible, and only after that we do mesh reconstruction. 3. It's such a great question! And to answer that we need to go a little deeper into the algorithm. Basically, a simple answer is that to run NeRFs and 3DGS, the framework input needs to be sequential, like the next frame has to be in timely order of the current frame, but if you take photos, the input will become non-sequential because we'll never guess where you took the next photo versus the current one, and this sequential image set is crucial for us to train the machine learning 4. I don't think wild camera movement has anything to do with the technical limitations, I guess it's just cooler to show it that way?
@@KIRI_Engine_App This is already out of the scope for most content creators here, but I just wanted to share these interesting aspects mentionned by Clive Fraser (Department of Infrastructure Engineering & CRCSI University of Melbourne, Australia) about the terminology often wrongfully used by most content creators (not you :)) “SLAM, SFM AND PHOTOGRAMMETRY: WHAT’S IN A NAME?” www.isprs.org/tc2-symposium2018/images/ISPRS-Interview-Fraser.pdf
Hahah good one :), But you know what, the reason why we wanted to create KIRI Engine is primarily because we hated 3D modeling lol, We are actually really bad at it
I've always been very optimistic about your software, please keep up the good work! Also, I wonder if there are any related paper links for me to study how to convert 3DGS into mesh?
Thank you so much Yichen!! Currently there's just SuGaR, but the paper is not that good :( The code that Chongjie is gonna open source soon isn't something that's entirely new, but a heavily engineered version based on existing algorithms.
The main problem with GS is that it isn’t scalable. If you attempt to scale it, the model breaks down, clearly revealing the individual splats. Additionally, in VR, GS appears like a low-resolution video. However, GS excels in handling complex and larger scenes.
Aaaawesome work, can't wait to try this!!! Really nice. This is a custom implementation of NeuS applied on top of Gaussian Splatting, right? Have you tried comparing it with Sugar for example?
I don't think the Mesh is created from 3DGS, but through Photogrammetry, then saved together with the splat data. Gaussian Splatting is a Rendering technique, no mesh is involved other than splats.
Hey dazrelixs! Thanks for the question! I actually made a tutorial video for scanning human head, the same methodology can also be used for scanning a full human body, hope it can help :) th-cam.com/video/bqN5IWSm2zE/w-d-xo.htmlsi=wDzTRAyEqFo0gfTu
Haha perfect!! I'm excited for your printers to come :) We had many resin printers here in our office as well, they work magical with KIRI Engine. One PRO tip, if you know how to clean up the mesh with Z Brush before printing, the final results will go to another level
Heyyy, I'm so sorry to hear that, we should have either refined or shipped all the Kickstarter orders years ago, could you please send us an email at contact@kiri-innov.com?
@@cekuhnen Hi there! Sometimes, the mesh can be problematic because it's 3DGS :( But may I take a closer look at the mesh? If you don't mind, please send the model to jack.w@kiri-innov.com, let me see what I can do to help you improve
Frankly speaking, it's not that hard, it's just that in our current version, the mesh quality from 3DGS isn't as great as photogrammetry, so we weren't sure whether the 3DGS would be useful at all. But thanks for the feedback, now we know PBR is the next right thing :)
Great video! Why doesnt kiri just make a 3dgs add on for blender? Theres one for ue5 but nothing good for blender. Id much rather use gaussian splat then mesh in some cases. Is there a reason most apps skip blender?
Hi! That's a great question! We are hiring talented artists to help us make 3DGS add-ons on Blender, I feel the exact same pain point. It's just that writing 3DGS add-ons for Blender requires some serious computer graphic skills, there isn't many people around that is capable of making 3DGS add-ons for Blender
Hey wrillywonka :) I know it's been 7 months haha, but I just bump into this old comment and wanted to let you know that thanks to your comment here, we've launched the Blender addon for 3DGS last month :) It is completely free and open-sourced, check it out: github.com/Kiri-Innovation/3dgs-render-blender-addon
Great question, and actually 3D Gaussian Splatting doesn't really need the dense point cloud. We actually tried generate 3DGS from dense point and we realized that the result with the dense point cloud is even worse than the sparst point
@@t.u.1885 Ohhh gotcha, well the answer is probably a no. Because to generate accurate dense point cloud, the object itself needs to have enough feature points. Here is the thing, if the object has enough features point, we don't really need 3DGS to generate dense point cloud because we already have photogrammetry which is more capable to do that :P But if the object doesn't have enough feature points, even if we can manage to let 3DGS generate dense point cloud, it won't be as accurate, so no use probably?
One question I have is whether the Kiri Engine uses Lidar data to generate the 3D Gaussian splats, because I always thought that with Lidar on the iPhone, there would be an extra layer of data that could help create a more accurate splat or at least speed up the training process by importing the tracking from the photos.
Hi Birugueta! Thanks for commenting and it's a good question! We did not use LiDAR data at this moment although like you said it indeed adds an extra layer of data (which is the camera pos data) to help align photos faster. However, we want to make sure more users (the majority who doesn't have an iPhone Pro, and also the Android users) can benefit from our app from day one, so we didn't put LiDAR layer to this 3DGS version
But it would be cool to have the extra option specifically for the phone that has this feature, so those who don’t have it can just keep it in conventional mode, right? That way, nobody is at a disadvantage and those who have access to the extra feature can achieve even better results.
@@pedrinhosari Totally agree! We are a small team of developers and researchers, it will take some time to perfect every aspect of the app, but we'll keep upgrading it :)
Honestly this is really cool but what I really want to see is a desktop version of KIRI that I can run locally on my machine. I notice the the scans are very fast but the quality does not match Polycam or other desktop apps like Photocatch. A desktop app would solve this
Hi there, thanks for commenting :) Actually, if you compare KIRI's photo scanning with other photogrammetry apps, you will actually realize our mesh quality is way better than other app and even comparable with professional software. It's just this new mesh-inclusive 3D Gaussian Splatting method doesn't have great mesh when comparing with photogrammetry, but hey, we are the only one here that can do mesh with 3DGS :)
BTW, not sure if you know, KIRI Engine also has a website version :) You can upload high-quality photo from DSLR camera to get professional-level 3D asset
I can give you a powerfull reason why 3DGS will never replace mesh, is simple, you can't edit GS, basically you get what does exist on the ay it exist, and I don't see any usefull way to edit it in the future, the only advantage of GS at the moment is to change the angle of something that already exist, so basically correct the camera position... you can do that with the real object, meanwhile with 3d mesh you can edit and transform the 3d object on whatever you want, combinate on many difrent ways, change texture, relight, edit... so, no I don't thing GS as it is will replace any 3d scan
It's very true zerozoneyt, thank you for the great comment! I have faith that 3DGS will gradually get editability based on how fast the research papers got published around 3DGS. And even here, we are working with many research teams trying to achieve 70% of what we can currently do with photogrammetry models. So right now, 3DGS doesn't really have many use cases like you said, but I think it might not be the case in the future
Hello, wonderful video and app ! I understand Gaussian splatting is better for 3D scanning and is associated with PLY file format. In 3D printing, would it be possible at some point to bypass OBJ/mesh format and work directly with Gaussian splatting format? Just curious and surely confused with this new tech :)
Hey NIX, great question, I'm sure many people as also confused with this. Unfortuantely, the short answer is probably a no. 3D Printing is stuck with mesh format because that's how it can actually fabricate the 3D model. But don't worry, we will continue to support and constantly update mesh formats so 3D printer user can benefit from the software
I remember I saw a user from our community on Discord shared their dental usecases, but I don't think there are many. Feel free to download KIRI Engine and try it out yourself :)
@@KIRI_Engine_App Hello, I'm in the group and I even purchase the paid pro version. I would like to know if it is possible to resize to the size of the real object.
@@lucasfigueiredo1871 Thank you so much for the support :) Yes, it is possible. One of our user - Kevin did a tutorial of how to get real-world dimension to the 3D model, here it is: th-cam.com/video/axdPIc6FqQU/w-d-xo.htmlsi=rDD9ccAeghOo1IMK&t=329
Thank you so much for the comment, OV support has been on our to-do list for a very long time :D Unfortunately we are a small team of engineers and there have been other matters that prioritized our plan. We are gonna speed up and try to bring OV support sooner
Heyyy NightWarp, you are not the only person who are confusing on how to scan a person properly, we are working on an in-depth tutorial video with TH-camrs
But in case you are in a hurry, here are some of my tips to scan a person's upper body: 1. Make sure the person doesn't move and make sure he/she stares at the same direction throughout the scanning 2. I find the best pose to keep a person stationary is to have his/her arms hugging his/her head in the back, this pose will make the person station 3. Make sure you are taking photos or videos quickly, the slower you take them, the more chances the person will move :) 4. Try the featureless object scan mode in KIRI Engine when scanning a person, it is more tolerant to the movement than photo scan mode
Hey NightWarp! I know it's been 7 months haha, but just wanna let you know that I've just made a tutorial on how to scan person :) Thought you'd like it th-cam.com/video/bqN5IWSm2zE/w-d-xo.htmlsi=u8yYwD3W9cAKi-ND
Hi there, thank you for the comment. Yeah I hate setting everything behind paywall either, it's just we tried to make feature free for once but it just drastically increased our cost and made it unsustainable :( Please send me an email at jack.w@kiri-innov.com, I'll see what I can do for you to let you test the feature
Thanks for the question Mojtaba :) Yes, this method generates both mesh and texture for the 3D Gaussian Splatting. But one thing to keep in mind that, in KIRI Engine app, you can also crop and edit the 3D Gaussian Splatting, but we haven't implemented the mesh generation algorithm in the cropping yet. So if you made crop on 3DGS, it won't have mesh for now :(
Hey Daniel, thanks for the question! Do you Rodin, the generative AI for 3D models? If so, KIRI and Rodin have quite some differences haha. In short, KIRI Engine is a 3D reconstruction product whereas Rodin is 3D generation product. The difference is that if you want to have an exact same replicate of the physical real-world objects, you need to use 3D reconstruction. But if you just want to bring 3D model from a sketch, doesn't need to be the exact same, then Rodin can be a good product to go
I installed your app then clicked 3DGS at the top and hit the + button to do a scan. There are a few options like photo scan, 3dgs(beta) and featureless object. I thought that since I clicked 3dGS at the top of the app, the photo scan I did would be a 3DGS format but instead it was a low poly photogrammetry scan. Does this mean I actually NEED to pay for pro to do a 3DGS format process?
Hey Monster Juiced! Unfortunately, for the 3DGS scan, you do need a PRO account. This is a very computationally intensive algorithm, and we regretfully do need to compensate for the server cost. However though, even for photogrammetry, the 3D model rendered in the app is low-poly but when you export we'll give you both low-poly and high-poly models in the folder for free. And if you would love to try the 3DGS scan, please feel free to email me at jack.w@kiri-innov.com. I'll see what I can do for you to help try the 3DGS scan :)
I'm just wondering if it's my app or not, but when I clean up the 3dgs on pc the obj option is blanked out even though on the original mesh it's there, is this something we just have to clean up in blender and can't do it through the app?
Hey Arial! This is our app issue. Currently, we don't have the algorithm to generate the mesh of an edited 3DGS. It's because we used the video you shot for both 3DGS and mesh, but when you edited the 3DGS, we can't use the original video any more and if we are going to use the 3DGS's own data for mesh generation, the texture will be super bad. We are still trying to find a way to work around, so please bear with us
Will there eventually be a way to convert our previous existing GS scans to mesh? I have some location scans that I cant return to, and it would be great to have them as mesh
Hi there! Thank you so much for the comment. I think it can be possible, although we wasn't really planning on that. If you use Discord, you can raise this up on our Discord server, and if others would like to see the same, we'll try our best to fulfill it, feel free to join us here: discord.gg/Jt2JHG7BAc
I'm confused... doesn't meshing basically throw out the advantages of gaussian splats? Converting splats to a mesh is a step backwards, no? It gives you standard compatibility, but you lose reflections, translucency, and performant point clouds. Why do this except to go back to 1995? I suppose the focus here is an alternative to photogrammetry. What am i missing?
Exactly!! I'm so glad to see you agree with my point as I have stated in the later part of this video :) Exactly like you said, by converting 3DGS to a mesh, it loses all the advantages of being 3DGS. But before 3DGS receive massive adaptation to major 3D creation pipeline (which we are working really hard on), in order for users to use 3DGS, we need a intermediate solution to convert it to mesh
Hey George, that's a very good question. Using OBJ has been our old habit haha, we probably need to change that. But from a technical point of view, OBJ is the most universal file format believe it or not. Because OBJ provides mesh and textures separately, this gives people more freedom to manipulate the data inside. And despite the recent hype in USD, it hasn't gotten full adaption to all 3D software
@@KIRI_Engine_App OBJ is just terrible.... and full adoption would require that startups like yours offer it. . . be part of the solution, not the problem please! BTW I do like this new ability as I do photogrammetry and see the need for simpler ways consumers can start capturing things.
@@GeorgeKrieger That's actually a good point. I don't know if this TH-cam comment area is the best place, but clearly, you are a 3D professional knows what you are doing, could you briefly let me know why you think USD is way better than OBJ? We know how to code, but we don't have a strong background as 3D artists :( So, the information would be really really helpful
Is it possible for me to get a month trial?, I am willing to put money towards the app but I need to see the pro results on my own files before putting the money into it.
Hey ParaMoto, thanks for commenting, but one month is too much and it'd be very unfair to other users. Please send me an email at jack.w@kiri-innov.com, I'll see what I can do to help you get a taste of the software before paying :)
@@KIRI_Engine_App Another question: can I upload a video on your website and have generate a mesh (with texture) from that? Do you use Neurangelo on your website? Now I see that only photos can be uploaded.
@@MirceaGoia Thanks for the question Mircea!! Depends on the scanning mode you use, for example if you use photo mode, then you can only upload photosets on the website and we use Photogrammetry algorithm to process the photosets; but if you use Featureless Object mode, we will ask you to upload a video instead of photos and we use a improved version of Neuralangelo for Featureless Object mode
Heyy, thanks for comment! Actually, KIRI Engine is also available on web browsers, feel free to go to www.kiriengine.app/web-version :) So it also works on Linux
Hi yousha! Thank you for the comment! I'm aware that there are free Gaussian Splatting options out there online, but we see our 3DGS scan as a one of the benefit for the KIRI Engine Pro users, by becoming a KIRI Engine Pro user, you not only get the GS, but also all kinds of powerful features including our mesh-inclusive 3DGS
Hi there, I respect your opinion, no offense taken. But we created KIRI Engine trying to solve the exact point you mentioned, you see, not many people can afford a thousand-dollar 3D reconstruction software, and definitely not many people have a powerful enough PC to run the software.
This guy's smile and pride in his product is amazing. I can't wait to try it (3dgs only)
Thank you so much! This comment just made my day better!!!
Thanks for a great video - is 3DGS coming to iPhone App Store UK? Thank you
I'm not gonna lie, but our development team is in the middle of trying to 3d model items and models, but because none of us can 3d Model(mainly because we're only 14😅) We in a bit of a hole. Becuase a good multi-player horror game needs good quality, and YOU my friend made my day, Thank you our team wishes you the best of luck for the future 🤩💪
Thank you so much haha! Good luck with the game :)
The biggest feature that would help photogrammetry workflows would be separating the speculative map from the basic material. or even creating materials from gaussian Spalatting. The way I see it, a Gaussian Spalt model is a “3D latent space” and a KI algorithm (probably even a cheap comvolutional approach) could explore to find out where primary reflection “move” over the surface and where ambient color stay stable even if perspective changes. The classic approaches to capture reflective maps in photogrammetry like photometric stereo or polarized flashes are very time consuming. It would be awsome if ai could “learn” where an object is “reflective” “translucent” or “refractive” by walking the 3d space of a Gaussian splat capture 🙏
Thank you for such a detailed comment, just for this comment, let me share some exciting news we are going to reveal soon :) So, basically, we are working on a noval inverse-rendering technique where you can not only 3D scan completely transparent and reflective objects and get mesh, if there is stuff inside the transparent object, for example, an ice cube inside a transparent bottle, this new technique can also reconstruct both the bottle and the ice cube inside, it's cool and I can't wait to show it to the public in the coming month
@@KIRI_Engine_App that sounds really exciting! I wonder if the turntable platform setup with fixed camera will work well for this. It is necessary for Gaussian splatting to move the camera relative to the object AND light sources? Or will the inverse rendering technique will cover stable camera, rotating object and stable lights (in relation to the camera, not the object) ?
No. You're Awesome. 😎
Great job by the way just needed to say that. As you likely know though the reasons 3d artists need a mesh and why mesh export is important the ability to at some point relight, change materials and fully integrate GS into a shot comprising of both kinds of assets in a believable way. Great first steps
That's a great comment, thank you Ethelbert! Yes, personally speaking (it's Jack here BTW), I'd be happy to see if we can have 3DGS achieve 70% of what photogrammetry can do by the end of next year, and for that goal, we are working to solve relighting, shadowing casting, physic etc problems on 3DGS
Dude! this is insane! i just scanned my house in 2 minutes! :D
Yoooo, I'm so glad to hear that :) Feel free to let us know if you have any questions
@@KIRI_Engine_App for sure I will! I’m a pro user now 😁
Oh wow. This is it! Now we can finally say adieu to photogrammetry . 3D modelling will become more fun.. Exporting as obj is cherry on top🙂. Thank you so much guys.🙇♂
Thank you so much man :) But I'm afraid that most people are gonna have to stay with photogrammetry for a little longer :P Because in this mesh-inclusive 3DGS version, the mesh quality is still not quite there comparing with photogrammetry, but we'll get there in the future
@@KIRI_Engine_App True. At least we have a hope that something is happening.
@@dreamzdziner8484 Yup, I'm quite optimistic about the future :)
theres still a long way to go.. but I'm glad it's here
this guy is so passionate about his work. love to see it. massive shoutout to the guys who make these powerful tools for artists
This is the future, superb work Kiri people
Thank youuuu ENOURMOUS my man!!
Hats off! Again. Crazy good. Thank you for the constant, high quality development of the Kiri engine. The future looks bright. #GaussianSplatting
Haha, thank you so much for the support, more to come, let's goooo
Idont know if you are using this SuGaR, but its nice to check about it: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction
th-cam.com/video/MAkFyWfiBQo/w-d-xo.html&ab_channel=AntoineGu%C3%A9don
Yes, we checked out SuGaR the first day it came, but it was way too bad to be used in an actual product. This algorithm is a heavily engineered version of NeuS, but it does the job much better :)
👏What a SPECTACULAR update!!! You and the team are freak'n kill'n it!!!
Thank you so much quietlyworking :DDD
You guys are incredible ! Thanks for all that you do !
I'm flattered haha, thank you so much for the incredible support!!
oh that is really gaming changing . i didn't talk enough Kiri engine in my yt channel , will do . thanks for your work and i love your videos as your app
Niiiice subbed :D
@@jackisbuildingkiri2997thank you, and again , amazing work with Kiri
Awesome! Installed and subscribed right away! As you say! Now lets hope the industry follows 👍💪🙏
Thank you for the sub and installation!! And let's see if the industry follows :D
Wow, this is amazing! I can't wait to try it out!
Yay!! Feel free to let us know if you have any questions :D
This is a great start, and I commend your efforts on outputing something useable. I think you know these objs are going to be dense meshes of hot garbage thought, and almost useless for any 3d modeling purposes (or worse than useless). The actual novel work begins if y'all choose to tackle the non-trivial challenge to output useful geometry, I know it might be too much, but good luck if y'all take this on, it would be a world first basically (Instant Meshes is like a 20% solution)
(of course 'useless' = recreating by hand would be faster than cleaning it up)
Hey there! That's a very insightful comment. I agree that even with mesh from 3DGS, it won't be as usable as other methods, but we just need to keep pushing the technology and adaption because if we don't, who will... If you are looking for more usable models, our photogrammetry and NeuS models are good for you, we have some serious stuff like quad mesh retopology and PBR generator to improve the usability, feel free to give it a try :D
Amazing stuff guys. Good work
Thank you so much !!!
There is two use cases (or technologies) in 3D reconstruction where Gaussian Splatters probably cannot compete with a mesh as 3d representation format but at the same time could trigger massive change in how nowadays they are used as in their visual quality (textures). They are Lidar mapping (large scale mapping) and 3D scanners. It is very difficult a Gaussian Splatters or it mesh can compete in precision and quality with such devices, used i.e. in survey or reverse engineering. In such cases the mesh you are able to get, can be used to aling the gaussian splatter representation with the lidar/3Dscanner mesh, and reproject the GS texture in their mesh.
import mesh into 3D software - create a texture that pipes color/texure from g.splatting renderer (capture an envelope of splats from around that mesh zone) - use the 3D mesh as a mask in 3D app.
would need to link cameras world coords geo/splat postitions etc in both 3D environments
screenspace masked texture (no UVs) .. "easy*" .. (but brute force) *easy for you genuises .. not me, sadly :D
@@LeeBrenton Oh wow, thank you so much for such a technical comment. We know how to write code, but we are not really 3D artist, so every time we see this kind of comment it made us understand a bit more about the realm. Please feel free to join us on Discord here, we'd love to talk with you more to see if we can further help with the pipeline: discord.gg/Jt2JHG7BAc
I have been waiting for this! So excited to try it!
Glad to hear that :) Please feel free to let us know if got any questions while trying it
i am following with curiosity you are doing great work
Thank you so much Burak :) We too, are excited about the journey ahead, let's see what's coming together :DDD
@@KIRI_Engine_App i will buy kiri engine pro to support you on this path. one thing I'm wondering is if there is a difference in the quality of the pro plan. and besides, you usually make videos on gaming platforms, kiri engini promises a very important future for e-commerce, I think it would be useful if you make videos about it.
For use cases where accurate measurements of digitalized objects are needed, do you see the future in:
A.) More and more methods emerging to convert the gausian splats into very accurate point clouds and mieshes (regarding geometrical and spatial fidelity), or
B.) more and more methods emerging to use the splats without conversion?
Do you see any chances that Gaussian splats will replace traditional photogrammetric models, for example for surface roughness estimation, object and 3D feature (like cracks in buildings or rocks) detection?
Great question GamingShiiep! If we are prioritizing accurate measurements, then mesh will always be the king, but if we are prioritizing visualization, then I do believe 3D Gaussian has a chance.
And I don't think 3D Gaussian Splatting will ever replace photogrammetry, but it will find its own place where photogrammetry is limited, for example, it's impossible to photogrammetry hair or thin leaves, but 3DGS can do it, so I think in the future creators get to use different representation within the same project to make the 3D scene more realistic
@@KIRI_Engine_App
Thanks for your answer!
I have a couple more questions (sorry!), but they might give you some ideas:
1) Assuming we use the same input frames/photos, hyperparameters and framework for the point cloud generation using photogrammetry and 3DGS, will we end up with the exact same positions of the points?
2.) Is the Mesh creation in Kiri "simply" a linear interpolation between the points or are other methods used to create more points by something like intercepts of various degrees of the Splats? Since things are going super fast and I'm lacking the math skills behind programming and enhancing such a solution (for now), I'm sure I don't need to keep such an idea for myself.
3.) When I assessed the quality of my photogrammetric model using video frames vs photos, the quality difference was huge and much better with much less photos. I strongly assume that it's because of noise, compression, and color depth etc. (I used my Google Pixel 7 pro cameras at 60 fps and 4k). Why does everybody in the context of NeRF and 3DGS always talk about using video then? For photogrammetry a higher number of video frames did not make up for the lack of quality compared to fewer photos (600 frames vs 60 photos. HUGE difference). Is that different for 3DGS or for NeRFs? If so, why?
4.) Why are most of these 3DGS and NeRF show case videos always using such wild camera movements and not just slow and smooth "camera rides"? I have absolutely no Idea why that would be. Also, isn't everything at the end rendered for interactive viewing on a screen? Isn't a photogrammetry model that you look at (either mesh or point cloud) also rendered to be seen on the screen?
@@GamingShiiep Wow, these are very well-thought-out questions! Let's do it:
1. Yes, we will end up with (almost) the same positions of the points if we use the same frameworks and parameters as inputs. The reason I wouldn't say the exact same is that even if you use the same input for photogrammetry two times, you won't get the exact same points. This is just how random the algorithm is
2. Yes and no, basically there are two steps of point cloud generation in our photogrammetry. There's an initial point cloud generation we call camera pos calculation using SFM techniques. After the first step, we will end up with sparse point cloud; Then we use the sparse point to extract more points from the photos, this step is called dense point cloud generation. After that, we use some mathematical appoint to align the points to make sure all point alignments are as smooth out as possible, and only after that we do mesh reconstruction.
3. It's such a great question! And to answer that we need to go a little deeper into the algorithm. Basically, a simple answer is that to run NeRFs and 3DGS, the framework input needs to be sequential, like the next frame has to be in timely order of the current frame, but if you take photos, the input will become non-sequential because we'll never guess where you took the next photo versus the current one, and this sequential image set is crucial for us to train the machine learning
4. I don't think wild camera movement has anything to do with the technical limitations, I guess it's just cooler to show it that way?
@@KIRI_Engine_App great answers! Thanks!
@@KIRI_Engine_App This is already out of the scope for most content creators here, but I just wanted to share these interesting aspects mentionned by Clive Fraser
(Department of Infrastructure Engineering & CRCSI
University of Melbourne, Australia) about the terminology often wrongfully used by most content creators (not you :))
“SLAM, SFM AND PHOTOGRAMMETRY: WHAT’S IN A NAME?”
www.isprs.org/tc2-symposium2018/images/ISPRS-Interview-Fraser.pdf
Thats nice! good work
Thanks :))))) Feel free to let me know if you got any questions :DDD
Ore you model everything yourself because you have skills :)
Hahah good one :), But you know what, the reason why we wanted to create KIRI Engine is primarily because we hated 3D modeling lol, We are actually really bad at it
Let's go! More features will be coming soon!
Noooo you took the first comment
LOL of course@@jackisbuildingkiri2997
I've always been very optimistic about your software, please keep up the good work! Also, I wonder if there are any related paper links for me to study how to convert 3DGS into mesh?
Thank you so much Yichen!! Currently there's just SuGaR, but the paper is not that good :( The code that Chongjie is gonna open source soon isn't something that's entirely new, but a heavily engineered version based on existing algorithms.
The main problem with GS is that it isn’t scalable. If you attempt to scale it, the model breaks down, clearly revealing the individual splats. Additionally, in VR, GS appears like a low-resolution video. However, GS excels in handling complex and larger scenes.
Great point cygnos! I actually didn't think of the scaling problem but now that you've mentioned, it is actually a huge issue with 3DGS.
This is awesome!
Thank you so much :)))
You guys rock! 🔥
OMG, thank you so much!! You rock too :P
Excellent work! I wish you so much success!
Thank you so much!!
Aaaawesome work, can't wait to try this!!! Really nice. This is a custom implementation of NeuS applied on top of Gaussian Splatting, right? Have you tried comparing it with Sugar for example?
I don't think the Mesh is created from 3DGS, but through Photogrammetry, then saved together with the splat data. Gaussian Splatting is a Rendering technique, no mesh is involved other than splats.
OMG thats cool!!
Really glad you liked it :)) More are coming
Hi! Awesome product but I would like to know if Kiri Engine can scan a full human body and which method to use to scan a full human body? Thanks!
Hey dazrelixs! Thanks for the question! I actually made a tutorial video for scanning human head, the same methodology can also be used for scanning a full human body, hope it can help :) th-cam.com/video/bqN5IWSm2zE/w-d-xo.htmlsi=wDzTRAyEqFo0gfTu
Awesome!
Haha, thank you so much :)
Amazing.....
Not all models are the same though....
Intelligence = Quad layout corresponding to deformation
UV's?
Hey Miles!! Not exactly sure what you question was :( Feel free to clarify that here :)
Can you make a crowdsourced competitor to google street view, but in 3d? It would be so cool to drive around and explore the world from a computer.
Actually it's totally possible :) We have a new product in the pipeline that can do this, stay tuned
I'm waiting for a large format resin 3d printer and then I'll try scanning with your app. I gotta change my phone too ;)
Haha perfect!! I'm excited for your printers to come :) We had many resin printers here in our office as well, they work magical with KIRI Engine. One PRO tip, if you know how to clean up the mesh with Z Brush before printing, the final results will go to another level
I never received my Kiri Kickstarter 3D scanner. 🤷🏻♂
Heyyy, I'm so sorry to hear that, we should have either refined or shipped all the Kickstarter orders years ago, could you please send us an email at contact@kiri-innov.com?
@@KIRI_Engine_App I received a beta test unit, but never the actual scanner. I can send an email. Thanks for your reply. :-)
@@kaassoufflex9023 Thank you so much! Please do send us an email at contact@kiri-innov.com
Can you also make a Gaussian Splash of a room and mesh that ?
LiDAR scans are so useless - nearly always distorted
I totally agree, LiDAR is not even close when it comes to usable accuracy. And yes, our 3DGS scan can scan a room and get a mesh from that
@@KIRI_Engine_App I did try to export as mesh from a 3DGS scan but the geometry is very problematic with surfaces being very wavy having holes etc.
@@cekuhnen Hi there! Sometimes, the mesh can be problematic because it's 3DGS :( But may I take a closer look at the mesh? If you don't mind, please send the model to jack.w@kiri-innov.com, let me see what I can do to help you improve
Congratulation for the update
Thank you so much!!! Let's gooo
If you could qlso have it generate PBR textures based on the GS capture, that would complete the whole thing.
Frankly speaking, it's not that hard, it's just that in our current version, the mesh quality from 3DGS isn't as great as photogrammetry, so we weren't sure whether the 3DGS would be useful at all. But thanks for the feedback, now we know PBR is the next right thing :)
Great video! Why doesnt kiri just make a 3dgs add on for blender? Theres one for ue5 but nothing good for blender. Id much rather use gaussian splat then mesh in some cases. Is there a reason most apps skip blender?
Hi! That's a great question! We are hiring talented artists to help us make 3DGS add-ons on Blender, I feel the exact same pain point. It's just that writing 3DGS add-ons for Blender requires some serious computer graphic skills, there isn't many people around that is capable of making 3DGS add-ons for Blender
@@KIRI_Engine_App i think that even more reason why you would makes lots of $$ if you made it first. Id pay for that for sure
@@KIRI_Engine_Appi hoppe you will find thoses talented artists. Can't wait for Blender AddOn...
Hey wrillywonka :) I know it's been 7 months haha, but I just bump into this old comment and wanted to let you know that thanks to your comment here, we've launched the Blender addon for 3DGS last month :) It is completely free and open-sourced, check it out: github.com/Kiri-Innovation/3dgs-render-blender-addon
@@TheTchaume Hey, Tchaume!! Guess what, after 7 months, it's here :D Free and open-sourced!!! blendermarket.com/products/3dgs-render-by-kiri-engine
Exciting feature! Is dense point cloud from 3DGS also coming soon?!
Great question, and actually 3D Gaussian Splatting doesn't really need the dense point cloud. We actually tried generate 3DGS from dense point and we realized that the result with the dense point cloud is even worse than the sparst point
@@KIRI_Engine_App My question was if we could generate dense point clouds from 3DGS.
@@t.u.1885 Ohhh gotcha, well the answer is probably a no. Because to generate accurate dense point cloud, the object itself needs to have enough feature points. Here is the thing, if the object has enough features point, we don't really need 3DGS to generate dense point cloud because we already have photogrammetry which is more capable to do that :P But if the object doesn't have enough feature points, even if we can manage to let 3DGS generate dense point cloud, it won't be as accurate, so no use probably?
One question I have is whether the Kiri Engine uses Lidar data to generate the 3D Gaussian splats, because I always thought that with Lidar on the iPhone, there would be an extra layer of data that could help create a more accurate splat or at least speed up the training process by importing the tracking from the photos.
Hi Birugueta! Thanks for commenting and it's a good question! We did not use LiDAR data at this moment although like you said it indeed adds an extra layer of data (which is the camera pos data) to help align photos faster. However, we want to make sure more users (the majority who doesn't have an iPhone Pro, and also the Android users) can benefit from our app from day one, so we didn't put LiDAR layer to this 3DGS version
But it would be cool to have the extra option specifically for the phone that has this feature, so those who don’t have it can just keep it in conventional mode, right? That way, nobody is at a disadvantage and those who have access to the extra feature can achieve even better results.
@@pedrinhosari Totally agree! We are a small team of developers and researchers, it will take some time to perfect every aspect of the app, but we'll keep upgrading it :)
Honestly this is really cool but what I really want to see is a desktop version of KIRI that I can run locally on my machine. I notice the the scans are very fast but the quality does not match Polycam or other desktop apps like Photocatch. A desktop app would solve this
Hi there, thanks for commenting :) Actually, if you compare KIRI's photo scanning with other photogrammetry apps, you will actually realize our mesh quality is way better than other app and even comparable with professional software. It's just this new mesh-inclusive 3D Gaussian Splatting method doesn't have great mesh when comparing with photogrammetry, but hey, we are the only one here that can do mesh with 3DGS :)
BTW, not sure if you know, KIRI Engine also has a website version :) You can upload high-quality photo from DSLR camera to get professional-level 3D asset
I can give you a powerfull reason why 3DGS will never replace mesh, is simple, you can't edit GS, basically you get what does exist on the ay it exist, and I don't see any usefull way to edit it in the future, the only advantage of GS at the moment is to change the angle of something that already exist, so basically correct the camera position... you can do that with the real object, meanwhile with 3d mesh you can edit and transform the 3d object on whatever you want, combinate on many difrent ways, change texture, relight, edit... so, no I don't thing GS as it is will replace any 3d scan
It's very true zerozoneyt, thank you for the great comment! I have faith that 3DGS will gradually get editability based on how fast the research papers got published around 3DGS. And even here, we are working with many research teams trying to achieve 70% of what we can currently do with photogrammetry models. So right now, 3DGS doesn't really have many use cases like you said, but I think it might not be the case in the future
Hello, wonderful video and app !
I understand Gaussian splatting is better for 3D scanning and is associated with PLY file format.
In 3D printing, would it be possible at some point to bypass OBJ/mesh format and work directly with Gaussian splatting format? Just curious and surely confused with this new tech :)
Hey NIX, great question, I'm sure many people as also confused with this. Unfortuantely, the short answer is probably a no. 3D Printing is stuck with mesh format because that's how it can actually fabricate the 3D model. But don't worry, we will continue to support and constantly update mesh formats so 3D printer user can benefit from the software
@@KIRI_Engine_App this will be great. Thank you KIRI Team 📸📸
I would like to know if there are reports of anyone using it for dental medical purposes? like plaster models of mouths?
I remember I saw a user from our community on Discord shared their dental usecases, but I don't think there are many. Feel free to download KIRI Engine and try it out yourself :)
@@KIRI_Engine_App Cool! I just made some models to learn from. Where can I find the community link on discord or telegram to join?
@@lucasfigueiredo1871 Awesomeee! Here is the link to our Discord server, see you on the other side: discord.gg/Jt2JHG7BAc
@@KIRI_Engine_App Hello, I'm in the group and I even purchase the paid pro version. I would like to know if it is possible to resize to the size of the real object.
@@lucasfigueiredo1871 Thank you so much for the support :) Yes, it is possible. One of our user - Kevin did a tutorial of how to get real-world dimension to the 3D model, here it is: th-cam.com/video/axdPIc6FqQU/w-d-xo.htmlsi=rDD9ccAeghOo1IMK&t=329
Have you looked into creating a Nvidia Omniverse extension? I would love to use it it OV for digital twins
Thank you so much for the comment, OV support has been on our to-do list for a very long time :D Unfortunately we are a small team of engineers and there have been other matters that prioritized our plan. We are gonna speed up and try to bring OV support sooner
any tips on scanning a person upper body? upper chst shoulder neck and head?
Heyyy NightWarp, you are not the only person who are confusing on how to scan a person properly, we are working on an in-depth tutorial video with TH-camrs
But in case you are in a hurry, here are some of my tips to scan a person's upper body:
1. Make sure the person doesn't move and make sure he/she stares at the same direction throughout the scanning
2. I find the best pose to keep a person stationary is to have his/her arms hugging his/her head in the back, this pose will make the person station
3. Make sure you are taking photos or videos quickly, the slower you take them, the more chances the person will move :)
4. Try the featureless object scan mode in KIRI Engine when scanning a person, it is more tolerant to the movement than photo scan mode
Hey NightWarp! I know it's been 7 months haha, but just wanna let you know that I've just made a tutorial on how to scan person :) Thought you'd like it th-cam.com/video/bqN5IWSm2zE/w-d-xo.htmlsi=u8yYwD3W9cAKi-ND
@@KIRI_Engine_App thank you!
I cant decide if I want to pay since the best features are behind a paywall
Hi there, thank you for the comment. Yeah I hate setting everything behind paywall either, it's just we tried to make feature free for once but it just drastically increased our cost and made it unsustainable :( Please send me an email at jack.w@kiri-innov.com, I'll see what I can do for you to let you test the feature
hello thanks for your fine app . since this update generate mesh does it include textures or i should use another method for scanning in this app ?
Thanks for the question Mojtaba :) Yes, this method generates both mesh and texture for the 3D Gaussian Splatting. But one thing to keep in mind that, in KIRI Engine app, you can also crop and edit the 3D Gaussian Splatting, but we haven't implemented the mesh generation algorithm in the cropping yet. So if you made crop on 3DGS, it won't have mesh for now :(
cropped 3DGS can not be exported in mesh format (the option is greyed). Why?
no original data to texture again.....
What do you think about Radin? And howbis it different then Kiri
Hey Daniel, thanks for the question! Do you Rodin, the generative AI for 3D models? If so, KIRI and Rodin have quite some differences haha. In short, KIRI Engine is a 3D reconstruction product whereas Rodin is 3D generation product. The difference is that if you want to have an exact same replicate of the physical real-world objects, you need to use 3D reconstruction. But if you just want to bring 3D model from a sketch, doesn't need to be the exact same, then Rodin can be a good product to go
I installed your app then clicked 3DGS at the top and hit the + button to do a scan. There are a few options like photo scan, 3dgs(beta) and featureless object. I thought that since I clicked 3dGS at the top of the app, the photo scan I did would be a 3DGS format but instead it was a low poly photogrammetry scan. Does this mean I actually NEED to pay for pro to do a 3DGS format process?
Hey Monster Juiced! Unfortunately, for the 3DGS scan, you do need a PRO account. This is a very computationally intensive algorithm, and we regretfully do need to compensate for the server cost. However though, even for photogrammetry, the 3D model rendered in the app is low-poly but when you export we'll give you both low-poly and high-poly models in the folder for free. And if you would love to try the 3DGS scan, please feel free to email me at jack.w@kiri-innov.com. I'll see what I can do for you to help try the 3DGS scan :)
I'm just wondering if it's my app or not, but when I clean up the 3dgs on pc the obj option is blanked out even though on the original mesh it's there, is this something we just have to clean up in blender and can't do it through the app?
Hey Arial! This is our app issue. Currently, we don't have the algorithm to generate the mesh of an edited 3DGS. It's because we used the video you shot for both 3DGS and mesh, but when you edited the 3DGS, we can't use the original video any more and if we are going to use the 3DGS's own data for mesh generation, the texture will be super bad. We are still trying to find a way to work around, so please bear with us
Will there eventually be a way to convert our previous existing GS scans to mesh?
I have some location scans that I cant return to, and it would be great to have them as mesh
Hi there! Thank you so much for the comment. I think it can be possible, although we wasn't really planning on that. If you use Discord, you can raise this up on our Discord server, and if others would like to see the same, we'll try our best to fulfill it, feel free to join us here: discord.gg/Jt2JHG7BAc
I'm confused... doesn't meshing basically throw out the advantages of gaussian splats? Converting splats to a mesh is a step backwards, no? It gives you standard compatibility, but you lose reflections, translucency, and performant point clouds. Why do this except to go back to 1995? I suppose the focus here is an alternative to photogrammetry. What am i missing?
Exactly!! I'm so glad to see you agree with my point as I have stated in the later part of this video :) Exactly like you said, by converting 3DGS to a mesh, it loses all the advantages of being 3DGS. But before 3DGS receive massive adaptation to major 3D creation pipeline (which we are working really hard on), in order for users to use 3DGS, we need a intermediate solution to convert it to mesh
Why use ancient OBJ format when USD is now available? Seems like an oversight that needs to be corrected . . .
Hey George, that's a very good question. Using OBJ has been our old habit haha, we probably need to change that. But from a technical point of view, OBJ is the most universal file format believe it or not. Because OBJ provides mesh and textures separately, this gives people more freedom to manipulate the data inside. And despite the recent hype in USD, it hasn't gotten full adaption to all 3D software
@@KIRI_Engine_App OBJ is just terrible.... and full adoption would require that startups like yours offer it. . . be part of the solution, not the problem please!
BTW I do like this new ability as I do photogrammetry and see the need for simpler ways consumers can start capturing things.
@@GeorgeKrieger That's actually a good point. I don't know if this TH-cam comment area is the best place, but clearly, you are a 3D professional knows what you are doing, could you briefly let me know why you think USD is way better than OBJ? We know how to code, but we don't have a strong background as 3D artists :( So, the information would be really really helpful
Is it possible for me to get a month trial?,
I am willing to put money towards the app but I need to see the pro results on my own files before putting the money into it.
Hey ParaMoto, thanks for commenting, but one month is too much and it'd be very unfair to other users. Please send me an email at jack.w@kiri-innov.com, I'll see what I can do to help you get a taste of the software before paying :)
Does Kiri engine exports the textures along with the OBJ?
Hey Mircea! Yes definitely! When you export the OBJ file, you will find the texture files along in the folder
@@KIRI_Engine_App Cool! Thanks!
@@KIRI_Engine_App Another question: can I upload a video on your website and have generate a mesh (with texture) from that? Do you use Neurangelo on your website? Now I see that only photos can be uploaded.
@@MirceaGoia Thanks for the question Mircea!! Depends on the scanning mode you use, for example if you use photo mode, then you can only upload photosets on the website and we use Photogrammetry algorithm to process the photosets; but if you use Featureless Object mode, we will ask you to upload a video instead of photos and we use a improved version of Neuralangelo for Featureless Object mode
Hi. Can have a pro trial option ?
Hi there :) I'll see what I can do haha, please email me at jack.w@kiri-innov.com
What a future we look into... i wish someone would hold my hand through the next 20 years until i reach my pension.
The color data can be export with texture in obj?
Yup, it comes with extra color maps
@@KIRI_Engine_App thank you! i will show your software to my company.👏👏
@@maritamachado8710 Woah, thank you so much!!! If you got any questions, feel free to shoot me an email at jack.w@kiri-innov.com
how long it takes to generate the mesh?
This process is integrated into the 3DGS training process, so it would take about 15mins to get the 3DGS, and the mesh is automatically included
thanks!@@KIRI_Engine_App
Needs a Linux desktop app.
Heyy, thanks for comment! Actually, KIRI Engine is also available on web browsers, feel free to go to www.kiriengine.app/web-version :) So it also works on Linux
Paid Gaussian Scenes
Hi yousha! Thank you for the comment! I'm aware that there are free Gaussian Splatting options out there online, but we see our 3DGS scan as a one of the benefit for the KIRI Engine Pro users, by becoming a KIRI Engine Pro user, you not only get the GS, but also all kinds of powerful features including our mesh-inclusive 3DGS
Luma AI was the first not you :)))
Actually that's not true, this is the first 3DGS that can generate mesh :), other apps use traditional methods to generate mesh
BAD interface (import pictures and you need to select manuall each and still nothing happens), expensive, no trial, no thank you
wont touch this due to dubs ... i would rather pay thousands and own the software i use than rent it
Hi there, I respect your opinion, no offense taken. But we created KIRI Engine trying to solve the exact point you mentioned, you see, not many people can afford a thousand-dollar 3D reconstruction software, and definitely not many people have a powerful enough PC to run the software.
Any spying with this app? R U willing to bet everything ion security while using this app?
Hi there, thanks for commenting :) We don't spy, haha. Spying doesn't help our company at all, we just want to bring better 3D tech to 3D creators
mesh is not "from 3dgs" .. it's made from the same source video as the 3dgs.. i doubt you can make a mesh "from a 3dgs" file
Actually we do make mesh from 3DGS file :P This approach is now open sourced, feel free to check out here: github.com/hugoycj