Love your videos. I've been in the photogrammetry world for 15 years, and know about the processes/workflows however you still have me super entertained and are teaching me new things (Polycam etC) thanks guys. Keep smiling, love it.
hey guys !!!! Just wanted to drop a comment. wanted to thank you both for making these awesome and very engaging videos ! The way you combine all the info , the humor , the video editing ! It's just so much fun to watch your videos from beginning to the end ! What a nice new breeze , the gaussian splatting.. so damn cool ! Keep up the good work , i watch all your videos the moment they come out. and i'm sure everybody else does to ! high five
Wow man ...it means so much..THANK YOU ... reading this comment will give us the energy and motivation to stay up and do another 3 videos for next week. We are glad that you guys are learning and at the same time getting entertained.
Seriously, the vibe in this video is so good! you guys really made the effort to turn a boring tutorial into a super fun video. I'm convinced, im taking my best lens, my D810 and im scanning something today, i can't wait!
Right dude?? Just imagine the moment these become properly optimised for video games ... Imagine the possibilities for some of the stuff we used to play!!
Sience when you turn splats into niagara particles you should be able to tell some particles to "die" so you can clean the splats that annoys your scene. Will be a plugin to do it in no time, im sure. Even you can make some kill particle volumes to crop it by hand in a controlled space. God, it seems easy in my head haha , I should try to do it.
Thank you brother 😍 We gotta do the Buggy ride together when you are in Dubai 🤟🏼 It was COD in real life vibes 😂 and yeah the splats are becoming more and more widely used now
So would the result be much cleaner if you guys did Still images? With the examples you showed there are lots of blurry areas, I'm just wondering how to get the best quality for actual usage in Unreal for example.
Can I animate the crop? I have a beautiful scene that looks like a painting and I would like to create an animation of this object "expanding". Another question, since each element is converted into a particle, can I animate the particles with small movements? (I know you're not the creators of the plugin ;) but maybe you've already been through these questions). BTW, thank you for the research and amazing content ;)
Well the plugin we use at the end of the video is converting these to niagara particles for unreal engine. But if you are looking for polygons then not at the moment
Well, i think if you go 360° around a thing you want to maximize vertical resolution, because horizontally you go around the thing anyways? Like with a panorama shot. You stitch the images horizontally anyways, so why not shoot vertically so you maximize resolution and thus detail?
@@PestifeRIf the thing is vertical (like a standing person), you shoot vertically. If the thing is horizontal (like the car here), you shoot horizontally.
This is amazing. Im wondering. If someone took some images from a movie of an actor, front, side, back etc, could a 3d gaussian splatt model head be created? Could it then be animated or converted to a metahuman? It would save tons of hours of work sculpting a likeness, or at least speed on the process. What do you guys think??
I'm still thinking this is not a production-ready method for most cases, but it is still so cool! I hope it will get better in the future so 3D modeling will be less time consuming.
Yooo! killed it again guys. can't wait to try it out with my Unreal Engine students. thanks heaps for the tips Just wondering have you tested the Gaussians Plugin with pathtracing in unreal?
hey guys! just wanted to drop a quick shoutout for this awesome video! Big thanks for sharing those super useful tips and tricks. One thing that could make it even better is if you mention the quality improvements you get from the old method - it's noticeably better!
Thank you very much for your words. Glad you enjoyed the video. For a fair quality comparison, we need to have both scans generated from the same input. It's a great idea and we are gonna do it in the next video
All I need to know, is this plug-in more stable than Luma’s ai plug-in? Every time I import a Gaussian splat created by Luma, on one of their demo environments UE5 keeps crashing 😢. Great video Ps. I prefer the Gaussian splats from Luma than polycam. In luma you can import as many photos as you want (or videos). In polycam they have a very low limit on their free version, and I haven’t been impressed enough so far with the results I get to confidently pay their annual subscription.
I enjoyed seeing this amazing technology. I'm curious if there are any manuals or reference materials for optimizing shutter speed, aperture width, etc. regarding camera shooting.
Wonder if you could somehow take out the "images" of the say the camels head when it's not still, and generate the angles you want actually and require to make the head appear to be still, and pipe the tweaked/fixed output "images" back into the g.splat process in order to then render out a wholly still camel 🐫 following those steps. As a sort of steps in between the gsplat and control "image/frames" within it that you don't want/want to tweak.
So responding to both concepts! First the idea of removing bad inputs is a great idea. We were just afraid of not having enough material so we submitted it all! But you are absolutely right, technically you should yield a cleaner scan that way. And as for modifying the splats, currently the tool in unreal is not extremely userfriendly and we are speaking of the crop tool on the plugin. The absolute best way would be to manully use a tool similar to a paint brush and edit the ellipsoids. Yes that would be awesome! A Game changer indeed within unreal engine!
Why not shoot stills in burst or interval mode? Then you also can set shutterspeed higher than 1/100 with a good camera app. You could shoot at 1/250 or 1/500 which will increase sharpness even more. Your still frames are now jpg or raw in 8K and if jpg the image sequence is already done. You can shoot this way just as quick as video and no need to extract stills from video. Also no video compression and optionally use raw stills as source. Raw stills in burst mode. Convert to the highest supported fileformat to convert. No compression, less artifacts, better splats. I haven’t tried myself but this would make sense to try:)
Not yet - just waiting for tomorrow and a trip to the city :) too cool!!! Is it also possible to cut everything off or just have the subject? @@badxstudio
Great question. Right now you can only crop it in unreal with this plugin. But we saw somebody just delete the ellipsoids in unity so that was crazy! @@veith3dclub
Hi there! I noticed a small mistake in the video regarding the relationship between aperture and depth of field. Just for clarity, a larger aperture (smaller f-number) results in a shallower depth of field, whereas a smaller aperture (larger f-number) gives a deeper depth of field. Keep up the great work with your videos!
really enjoyed the video, and it was quite interesting to see what to expect from this technique. I think it's not quite there yet, but you managed to get some good results nonetheless!
Thank you brother Glad that you enjoyed the vid. Definitely all too experimental at the moment but it's surely going to substantially improve overtime!
The unity or blender addons are free i believe but we tested the blender one and it didnt work on windows .. only worked on linux as far as we are concerned! They are apparently working on it
why can't you cast and or receive shadows from other non gsplat objects in your scene. it's cool, but otherwise if the other objects don't have shadows or get reflected then it's like, why did we even import it into unreal. the re-lighting is ok though. seems more of a tint and exposure thing.
What would be wild, would be to take an AI generated image or sequence and make that 3D using this method. Then the AI would be able to not only generate the entire image sequence, but also generate the entire 3D environment. Possible?
Sooo...if I shoot a 30s video and convert it into an image sequence, it means I will end up having 720 images. As the maximum images you can upload in polycam is 200...I guess you have to choose between those 720, dont you?..
You are right. You don't need every frame as an image for creating a Gsplat. make sure there are differences between images. We use ImageMagik and probably only taking 2-4 image per second only
@@badxstudio great!!..im trying GS on polycam today!. Thanks for the info and for the content on your channel. I see now the maximum q of images is 250..yessss, the more, the better 🙂
what do you mean 'a specific altered version of Unreal'? It's just unreal 5, with the metashoot plugin :-D. lol. Anyway, great video's guys. Keep going
Omg thanks to you guys pointing it out, I discovered that there were two ghosts behind me watching the vid as well, and yes those mfers hadn't subscribed 🤦
Cant wait untill it has some real applications on tv/movie industry, for now it just cant generate any useful mesh that could be used in houdini/redshift/Vray/arnold etc, everything would have to be re-topologized anyways, its great technology but still waiting for some real life practical use that could come handy on cgi work. We use photogrammetry a lot but everything has to be retopologized so its optimized and looks great on physical based render engines