Wow - great video. I was just looking at the previous fork and was wondering if an actual implementation would be out soon, so this video is perfectly timed. Out of curiosity - have you worked out how to train only a specific area (say within a bounding box) to increase performance? I have cleaned/cropped my sparse point cloud, however it still seems to train far "outside" of the sparse point cloud, and I feel performance could be significantly increased if it were to only train within an area of interest. Thanks for all your time you have put into these videos!
Glad you found it helpful! I am not sure if it is possible to just train within the cropping box of a scene. That would be a great question for the team on Discord.
Thank you, Jonathan. Great video, as always! You mentioned that the licensing differs from Inria's original 3DGS? Does this mean that Nerfstudio uses its own original code?
thank you for a very intuitive guide on this man!! you made it so easy to understand... but I can make it to train if I do the nerfacto.. but not the gaussian-splat are those 2 different?
Those two are different training models. Ns-Train Nerfacto trains a NeRF dataset, ns-train gaussian-splatting trains a gaussian splat dataset. Both are radiance fields, but completely different output models. If you can't get ns-train gaussian-splatting to run, maybe you have an old install of Nerfstudio. You need to update as of January 1ish. You know if you can train gaussian splats if you type in "ns-train --help" and see "Gaussian-Splatting" as a training model choice.
@@thenerfguru thanks man !! I was thinking maybe my GPU is not up to the task.. I am using 12gb VRAM RTX 2080ti.. is it enough? or do i need rtx 3000 / 4000 series?
Hello there. It's a great video on how to use it, especially with installation and this one. Can we export a mesh from Gaussian splitting? There may be an option in the viewer under the point cloud. I haven't used it yet, so I reviewed your tutorial. Thanks.
I tried training gaussian-splatting on a RealityCapture project, but it seems like the gaussian-splatting only support COLMAP and when processing the Reality Capture .csv file you get a regular nerfstudio project without the colmap files. Have you had any success with Reality Capture?
I have not tried any other image prep method. This may solve your issue though: radiancefields.com/kapture-convert-for-faster-image-poses-from-realitycapture/
Yes! There are several plugins for UE5, just look into the marketplace. I also have a tutorial for Unity if you look in my channel. Here is also a good list of viewers: github.com/MrNeRF/awesome-3D-gaussian-splatting?tab=readme-ov-file#game-engines
Dear author, hello. Regarding your Gaussian pasting, I would like to ask, actually, I want to use it to export 3D models. Yes, I see that the generated effect is very realistic, and I want to export the characters or objects in the scene separately to 3D formats such as OBJ for my own use. Can I do this?
HI Thank you very much for the content it is amazing! I got an error when I am trying to train the data set : ImportError: cannot import name 'csrc' from 'gsplat' (C:\Users\XXXX\anaconda3\envs erfstudio\lib\site-packages\gsplat\__init__.py) As the name was changed for "splatfacto" I changed the for this statement : "ns-train splatfacto --data data/nerfstudio/poster" Do you know how to solve it? Thank you very much in advance.
(nerfstudio) C:\Users\igota erfstudio>ns-download-data nerfstudio --capture-name=poster 'ns-download-data' is not recognized as an internal or external command, operable program or batch file. I AM GETTING ERROR LIKE THIS, WHAT TO DO?
Hi! I was testing the data is provide but I get several errors when training, is it any way to share with you the log I get when processing? is there any channel where to share all these errors? This is an example: C:\Users\Jorge\miniconda3\envs erfstudio\lib\site-packages\torch\utils\cpp_extension.p y:383: UserWarning: Error checking compiler version for cl: [WinError 2] El sistema no puede encontrar el archivo especificado warnings.warn(f'Error checking compiler version for {compiler}: {error}') Thanks !
Hi, I installed Anaconda (not Miniconda) and when I put ns-train gaussian-splatting --data data/nerfstudio/poster command it says Parsing error. What could be a problem?
Hello ! First of all, many thanks for your very usefull videos 🙏 It's super easy to understand and really efficient. Well done :) I never had any trouble that i didn't manage to resolve with your advices or internet. But this time I have an apparently easy one that I don't find how to solve... I want to train my own datas, and when I try to do this command : ns-process-data {images, video} --data {DATA_PATH} --output-dir {PROCESSED_DATA_DIR} I have this error : Error : Video does not exist : data....Etc... But the video is in the right place "nerfstudio/data/...." and I keep trying to put every path possible in the environment variables without any success .... Maybe someone already meet this issue ? Look simple but annoying haha
@@thenerfguru No space and I already try several times with different naming to drag the file, it didn't work unfortunately. The thing is that I can see a first "images" folder created just next to my video file into the good folder when I press enter...so it looks like it understand the path... but can't see the video file in it... it is a short .mov and I can see it when I enter the "dir" command....(it doesn't work with images either) I finally decided to start from scratch, uninstall everything and back at it again. Works fine :) Who knows... Thanks for your time and the amazing community of GitHub :)
ns-train gaussian-splatting --data data/nerfstudio/poster not working above command gaussian-splatting but it works for ns-train nerfacto --data data/nerfstudio/poster whats the issue please suggest solution
are there any standard formats that gsplat renderers accept, e.g can the trained nerfstudio model checkpoint be used as is or are more intermediate export steps needed? export to .ply springs to mind @@thenerfguru
I haven't tried it yet myself. Looks like it trains on dense point cloud data as opposed to sparse. If it works out well for me, I will eventually make a video.
I really enjoy your videos! Keep up the good work my friend.
Wow - great video. I was just looking at the previous fork and was wondering if an actual implementation would be out soon, so this video is perfectly timed. Out of curiosity - have you worked out how to train only a specific area (say within a bounding box) to increase performance? I have cleaned/cropped my sparse point cloud, however it still seems to train far "outside" of the sparse point cloud, and I feel performance could be significantly increased if it were to only train within an area of interest. Thanks for all your time you have put into these videos!
Glad you found it helpful! I am not sure if it is possible to just train within the cropping box of a scene. That would be a great question for the team on Discord.
@@muhammadichsan1745 Yes, depending on the scene, masking might work! Ask around in the Discord, people have experience with masking.
example data (ns-download-data etc) is offline.
Thank you, Jonathan. Great video, as always!
You mentioned that the licensing differs from Inria's original 3DGS?
Does this mean that Nerfstudio uses its own original code?
Yes, that is my understanding.
Thank you so much Jonathan !
Are you considering doing a video on how to use nerfstudio in google collab?
thank you for a very intuitive guide on this man!! you made it so easy to understand... but I can make it to train if I do the nerfacto.. but not the gaussian-splat are those 2 different?
Those two are different training models. Ns-Train Nerfacto trains a NeRF dataset, ns-train gaussian-splatting trains a gaussian splat dataset. Both are radiance fields, but completely different output models. If you can't get ns-train gaussian-splatting to run, maybe you have an old install of Nerfstudio. You need to update as of January 1ish. You know if you can train gaussian splats if you type in "ns-train --help" and see "Gaussian-Splatting" as a training model choice.
@@thenerfguru thanks man !!
I was thinking maybe my GPU is not up to the task.. I am using 12gb VRAM RTX 2080ti.. is it enough? or do i need rtx 3000 / 4000 series?
Made it work just now!!! with the recent update of gspalt I just typed ns-train splatfacto and it worked now!!! thanks for guidance my man!!!
Hello there. It's a great video on how to use it, especially with installation and this one. Can we export a mesh from Gaussian splitting? There may be an option in the viewer under the point cloud. I haven't used it yet, so I reviewed your tutorial. Thanks.
Nice one ! Thanks 👏🏻👌🏻
You’re welcome!
Amazing as always NeRF Guru
Thank you so much 😀
I tried training gaussian-splatting on a RealityCapture project, but it seems like the gaussian-splatting only support COLMAP and when processing the Reality Capture .csv file you get a regular nerfstudio project without the colmap files. Have you had any success with Reality Capture?
I have not tried any other image prep method. This may solve your issue though: radiancefields.com/kapture-convert-for-faster-image-poses-from-realitycapture/
Hi! Thans!
Can we use 360 videos for training?
Hi, new here and I just subbed! Is there a way to get your splats into softwares like Unreal Engine for example?
Yes! There are several plugins for UE5, just look into the marketplace. I also have a tutorial for Unity if you look in my channel. Here is also a good list of viewers: github.com/MrNeRF/awesome-3D-gaussian-splatting?tab=readme-ov-file#game-engines
Thanks, man!
You’re welcome!
Dear author, hello. Regarding your Gaussian pasting, I would like to ask, actually, I want to use it to export 3D models. Yes, I see that the generated effect is very realistic, and I want to export the characters or objects in the scene separately to 3D formats such as OBJ for my own use. Can I do this?
That’s not the intended goal for this software. You will have a better result sticking to traditional photogrammetry.
thank you very much
Wooooooo! I've been waiting for this video!
Yes! Now’s the perfect time to give it a go too. They worked out initial bugs. It runs really well.
HI Thank you very much for the content it is amazing!
I got an error when I am trying to train the data set :
ImportError: cannot import name 'csrc' from 'gsplat' (C:\Users\XXXX\anaconda3\envs
erfstudio\lib\site-packages\gsplat\__init__.py)
As the name was changed for "splatfacto" I changed the for this statement :
"ns-train splatfacto --data data/nerfstudio/poster"
Do you know how to solve it?
Thank you very much in advance.
As I understand instaling COLMAP in your case just paste link in variables ? I tweak diffirent methonds of csutom instalation few hours ...
(nerfstudio) C:\Users\igota
erfstudio>ns-download-data nerfstudio --capture-name=poster
'ns-download-data' is not recognized as an internal or external command,
operable program or batch file.
I AM GETTING ERROR LIKE THIS, WHAT TO DO?
Out of memory when training a nerf on my own data but not on any of the demo data 🤷
Is it now possible to train 360 equi. Images with nerfstudio gaussiansplats?
Not natively. This is one workaround: ua-cam.com/video/LQNBTvgljAw/v-deo.html
I used my 360 Video with with ns-train splatfacto and it worked beautifully. Insta 360 Pro 2
My output are "Sparse_pc.ply" instead point_cloud.ply" is that normal or did i screw something up?
hi thanks for this video!
Is it available in colab with same way?
Is there others option to colmap ? its awfully slow and buggy on my machine :(
my very first line of code is giving an error (i followed the installation guide:( )
Hi! I was testing the data is provide but I get several errors when training, is it any way to share with you the log I get when processing? is there any channel where to share all these errors?
This is an example:
C:\Users\Jorge\miniconda3\envs
erfstudio\lib\site-packages\torch\utils\cpp_extension.p
y:383: UserWarning: Error checking compiler version for cl: [WinError 2] El sistema no
puede encontrar el archivo especificado
warnings.warn(f'Error checking compiler version for {compiler}: {error}')
Thanks !
My guess is that PyTorch did not compile during installation.
The Nerfstudio discord channel is great for these questions.
Hi, I installed Anaconda (not Miniconda) and when I put ns-train gaussian-splatting --data data/nerfstudio/poster command it says Parsing error. What could be a problem?
The newest version of the code has changed the command. Try ns-train splatfacto (gaussian-splatting has been deprecated)
Where did you create that transforms json file
Does anyone know if there is a forum about NeRF and 3DGS?
There are several Discord channels
How do you create that transforms.json file ?
thank you
Hello ! First of all, many thanks for your very usefull videos 🙏
It's super easy to understand and really efficient. Well done :)
I never had any trouble that i didn't manage to resolve with your advices or internet. But this time I have an apparently easy one that I don't find how to solve...
I want to train my own datas, and when I try to do this command : ns-process-data {images, video} --data {DATA_PATH} --output-dir {PROCESSED_DATA_DIR}
I have this error : Error : Video does not exist : data....Etc... But the video is in the right place "nerfstudio/data/...." and I keep trying to put every path possible in the environment variables without any success ....
Maybe someone already meet this issue ? Look simple but annoying haha
Could a folder have a space in it? Try dragging the file into command line and it should put the right path in there for you.
@@thenerfguru No space and I already try several times with different naming to drag the file, it didn't work unfortunately. The thing is that I can see a first "images" folder created just next to my video file into the good folder when I press enter...so it looks like it understand the path... but can't see the video file in it... it is a short .mov and I can see it when I enter the "dir" command....(it doesn't work with images either)
I finally decided to start from scratch, uninstall everything and back at it again. Works fine :) Who knows...
Thanks for your time and the amazing community of GitHub :)
ns-train gaussian-splatting --data data/nerfstudio/poster
not working above command gaussian-splatting
but it works for ns-train nerfacto --data data/nerfstudio/poster
whats the issue please suggest solution
Looks like they changed the name of the training method. Try name-train splatfacto
how can i insert my product into an existing video, this is for promotional ads
Where can we use gaussian splats other than ue5 and unity?
Here is a list of viewers you can integrate it with: github.com/MrNeRF/awesome-3D-gaussian-splatting?tab=readme-ov-file#viewers
@@thenerfguru i see it says blender. Have you tried it with blender?
@@wrillywonka1320 I have not. I have very little blender experience. I may give it a shot though. If it works out, I will make a video.
once trained how do you share the result?
How would you like to share your results? You can render a video, add it to a game engine environment, or export the file.
Here is a list of viewers you can showcase your results with: github.com/MrNeRF/awesome-3D-gaussian-splatting?tab=readme-ov-file#viewers
are there any standard formats that gsplat renderers accept, e.g can the trained nerfstudio model checkpoint be used as is or are more intermediate export steps needed? export to .ply springs to mind @@thenerfguru
Will you also do a tutorial on trilinear point splatting?
ua-cam.com/video/Nw4A1tIcErQ/v-deo.html
I haven't tried it yet myself. Looks like it trains on dense point cloud data as opposed to sparse. If it works out well for me, I will eventually make a video.