Bernhard Kerbl*, Georgios Kopanas*, Thomas Leimkühler, George Drettakis (* indicates equal contribution)
| Webpage | Full Paper | Video | Other GRAPHDECO Publications | FUNGRAPH project page |
| T&T+DB COLMAP (650MB) | Pre-trained Models (14 GB) | Viewers for Windows (60MB) | Evaluation Images (7 GB) |

This repository contains the official authors implementation associated with the paper "3D Gaussian Splatting for Real-Time Radiance Field Rendering", which can be found here. We further provide the reference images used to create the error metrics reported in the paper, as well as recently created, pre-trained models.

Abstract: Radiance Field methods have recently revolutionized novel-view synthesis of scenes captured with multiple photos or videos. However, achieving high visual quality still requires neural networks that are costly to train and render, while recent faster methods inevitably trade off speed for quality. For unbounded and complete scenes (rather than isolated objects) and 1080p resolution rendering, no current method can achieve real-time display rates. We introduce three key elements that allow us to achieve state-of-the-art visual quality while maintaining competitive training times and importantly allow high-quality real-time (≥ 30 fps) novel-view synthesis at 1080p resolution. First, starting from sparse points produced during camera calibration, we represent the scene with 3D Gaussians that preserve desirable properties of continuous volumetric radiance fields for scene optimization while avoiding unnecessary computation in empty space; Second, we perform interleaved optimization/density control of the 3D Gaussians, notably optimizing anisotropic covariance to achieve an accurate representation of the scene; Third, we develop a fast visibility-aware rendering algorithm that supports anisotropic splatting and both accelerates training and allows realtime rendering. We demonstrate state-of-the-art visual quality and real-time rendering on several established datasets.

@Article{kerbl3Dgaussians,
      author       = {Kerbl, Bernhard and Kopanas, Georgios and Leimk{\"u}hler, Thomas and Drettakis, George},
      title        = {3D Gaussian Splatting for Real-Time Radiance Field Rendering},
      journal      = {ACM Transactions on Graphics},
      number       = {4},
      volume       = {42},
      month        = {July},
      year         = {2023},
      url          = {https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/}
}

This research was funded by the ERC Advanced grant FUNGRAPH No 788065. The authors are grateful to Adobe for generous donations, the OPAL infrastructure from Université Côte d’Azur and for the HPC resources from GENCI–IDRIS (Grant 2022-AD011013409). The authors thank the anonymous reviewers for their valuable feedback, P. Hedman and A. Tewari for proofreading earlier drafts also T. Müller, A. Yu and S. Fridovich-Keil for helping with the comparisons.

We have limited resources for maintaining and updating the code. However, we have added a few new features since the original release that are inspired by some of the excellent work many other researchers have been doing on 3DGS. We will be adding other features within the ability of our resources.

Update of October 2024: We integrated training speed acceleration and made it compatible with depth regularization, anti-aliasing and exposure compensation. We have enhanced the SIBR real time viewer by correcting bugs and adding features in the Top View that allows visualization of input and user cameras.

Update of Spring 2024: Orange Labs has kindly added OpenXR support for VR viewing.

Jonathan Stephens made a fantastic step-by-step tutorial for setting up Gaussian Splatting on your machine, along with instructions for creating usable datasets from videos. If the instructions below are too dry for you, go ahead and check it out here.

User camenduru was kind enough to provide a Colab template that uses this repo's source (status: August 2023!) for quick and easy access to the method. Please check it out here.

The repository contains submodules, thus please check it out with

# SSH
git clone [email protected]:graphdeco-inria/gaussian-splatting.git --recursive

or

# HTTPS
git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive

The codebase has 4 main components:

• A PyTorch-based optimizer to produce a 3D Gaussian model from SfM inputs
• A network viewer that allows to connect to and visualize the optimization process
• An OpenGL-based real-time viewer to render trained models in real-time.
• A script to help you turn your own images into optimization-ready SfM data sets

The components have different requirements w.r.t. both hardware and software. They have been tested on Windows 10 and Ubuntu Linux 22.04. Instructions for setting up and running each of them are found in the sections below.

The optimizer uses PyTorch and CUDA extensions in a Python environment to produce trained models.

• CUDA-ready GPU with Compute Capability 7.0+
• 24 GB VRAM (to train to paper evaluation quality)
• Please see FAQ for smaller VRAM configurations

• Conda (recommended for easy setup)
• C++ Compiler for PyTorch extensions (we used Visual Studio 2019 for Windows)
• CUDA SDK 11 for PyTorch extensions, install after Visual Studio (we used 11.8, known issues with 11.6)
• C++ Compiler and CUDA SDK must be compatible

Our default, provided install method is based on Conda package and environment management:

SET DISTUTILS_USE_SDK=1 # Windows only
conda env create --file environment.yml
conda activate gaussian_splatting

Please note that this process assumes that you have CUDA SDK 11 installed, not 12. For modifications, see below.

Tip: Downloading packages and creating a new environment with Conda can require a significant amount of disk space. By default, Conda will use the main system hard drive. You can avoid this by specifying a different package download location and an environment on a different drive:

conda config --add pkgs_dirs <Drive>/<pkg_path>
conda env create --file environment.yml --prefix <Drive>/<env_path>/gaussian_splatting
conda activate <Drive>/<env_path>/gaussian_splatting

If you can afford the disk space, we recommend using our environment files for setting up a training environment identical to ours. If you want to make modifications, please note that major version changes might affect the results of our method. However, our (limited) experiments suggest that the codebase works just fine inside a more up-to-date environment (Python 3.8, PyTorch 2.0.0, CUDA 12). Make sure to create an environment where PyTorch and its CUDA runtime version match and the installed CUDA SDK has no major version difference with PyTorch's CUDA version.

Some users experience problems building the submodules on Windows (cl.exe: File not found or similar). Please consider the workaround for this problem from the FAQ.

To run the optimizer, simply use

python train.py -s <path to COLMAP or NeRF Synthetic dataset>

Note that similar to MipNeRF360, we target images at resolutions in the 1-1.6K pixel range. For convenience, arbitrary-size inputs can be passed and will be automatically resized if their width exceeds 1600 pixels. We recommend to keep this behavior, but you may force training to use your higher-resolution images by setting -r 1.

The MipNeRF360 scenes are hosted by the paper authors here. You can find our SfM data sets for Tanks&Temples and Deep Blending here. If you do not provide an output model directory (-m), trained models are written to folders with randomized unique names inside the output directory. At this point, the trained models may be viewed with the real-time viewer (see further below).

By default, the trained models use all available images in the dataset. To train them while withholding a test set for evaluation, use the --eval flag. This way, you can render training/test sets and produce error metrics as follows:

python train.py -s <path to COLMAP or NeRF Synthetic dataset> --eval # Train with train/test split
python render.py -m <path to trained model> # Generate renderings
python metrics.py -m <path to trained model> # Compute error metrics on renderings

If you want to evaluate our pre-trained models, you will have to download the corresponding source data sets and indicate their location to render.py with an additional --source_path/-s flag. Note: The pre-trained models were created with the release codebase. This code base has been cleaned up and includes bugfixes, hence the metrics you get from evaluating them will differ from those in the paper.

python render.py -m <path to pre-trained model> -s <path to COLMAP dataset>
python metrics.py -m <path to pre-trained model>

We further provide the full_eval.py script. This script specifies the routine used in our evaluation and demonstrates the use of some additional parameters, e.g., --images (-i) to define alternative image directories within COLMAP data sets. If you have downloaded and extracted all the training data, you can run it like this:

python full_eval.py -m360 <mipnerf360 folder> -tat <tanks and temples folder> -db <deep blending folder>

In the current version, this process takes about 7h on our reference machine containing an A6000. If you want to do the full evaluation on our pre-trained models, you can specify their download location and skip training.

python full_eval.py -o <directory with pretrained models> --skip_training -m360 <mipnerf360 folder> -tat <tanks and temples folder> -db <deep blending folder>

If you want to compute the metrics on our paper's evaluation images, you can also skip rendering. In this case it is not necessary to provide the source datasets. You can compute metrics for multiple image sets at a time.

python full_eval.py -m <directory with evaluation images>/garden ... --skip_training --skip_rendering

We provide two interactive viewers for our method: remote and real-time. Our viewing solutions are based on the SIBR framework, developed by the GRAPHDECO group for several novel-view synthesis projects.

• OpenGL 4.5-ready GPU and drivers (or latest MESA software)
• 4 GB VRAM recommended
• CUDA-ready GPU with Compute Capability 7.0+ (only for Real-Time Viewer)

• Visual Studio or g++, not Clang (we used Visual Studio 2019 for Windows)
• CUDA SDK 11, install after Visual Studio (we used 11.8)
• CMake (recent version, we used 3.24)
• 7zip (only on Windows)

We provide pre-built binaries for Windows here. We recommend using them on Windows for an efficient setup, since the building of SIBR involves several external dependencies that must be downloaded and compiled on-the-fly.

If you cloned with submodules (e.g., using --recursive), the source code for the viewers is found in SIBR_viewers. The network viewer runs within the SIBR framework for Image-based Rendering applications.

CMake should take care of your dependencies.

cd SIBR_viewers
cmake -Bbuild .
cmake --build build --target install --config RelWithDebInfo

You may specify a different configuration, e.g. Debug if you need more control during development.

You will need to install a few dependencies before running the project setup.

# Dependencies
sudo apt install -y libglew-dev libassimp-dev libboost-all-dev libgtk-3-dev libopencv-dev libglfw3-dev libavdevice-dev libavcodec-dev libeigen3-dev libxxf86vm-dev libembree-dev
# Project setup
cd SIBR_viewers
cmake -Bbuild . -DCMAKE_BUILD_TYPE=Release # add -G Ninja to build faster
cmake --build build -j24 --target install

Backwards compatibility with Focal Fossa is not fully tested, but building SIBR with CMake should still work after invoking

git checkout fossa_compatibility

The SIBR interface provides several methods of navigating the scene. By default, you will be started with an FPS navigator, which you can control with W, A, S, D, Q, E for camera translation and I, K, J, L, U, O for rotation. Alternatively, you may want to use a Trackball-style navigator (select from the floating menu). You can also snap to a camera from the data set with the Snap to button or find the closest camera with Snap to closest. The floating menues also allow you to change the navigation speed. You can use the Scaling Modifier to control the size of the displayed Gaussians, or show the initial point cloud.

After extracting or installing the viewers, you may run the compiled SIBR_remoteGaussian_app[_config] app in <SIBR install dir>/bin, e.g.:

./<SIBR install dir>/bin/SIBR_remoteGaussian_app

The network viewer allows you to connect to a running training process on the same or a different machine. If you are training on the same machine and OS, no command line parameters should be required: the optimizer communicates the location of the training data to the network viewer. By default, optimizer and network viewer will try to establish a connection on localhost on port 6009. You can change this behavior by providing matching --ip and --port parameters to both the optimizer and the network viewer. If for some reason the path used by the optimizer to find the training data is not reachable by the network viewer (e.g., due to them running on different (virtual) machines), you may specify an override location to the viewer by using -s <source path>.

After extracting or installing the viewers, you may run the compiled SIBR_gaussianViewer_app[_config] app in <SIBR install dir>/bin, e.g.:

./<SIBR install dir>/bin/SIBR_gaussianViewer_app -m <path to trained model>

It should suffice to provide the -m parameter pointing to a trained model directory. Alternatively, you can specify an override location for training input data using -s. To use a specific resolution other than the auto-chosen one, specify --rendering-size <width> <height>. Combine it with --force-aspect-ratio if you want the exact resolution and don't mind image distortion.

To unlock the full frame rate, please disable V-Sync on your machine and also in the application (Menu → Display). In a multi-GPU system (e.g., laptop) your OpenGL/Display GPU should be the same as your CUDA GPU (e.g., by setting the application's GPU preference on Windows, see below) for maximum performance.

In addition to the initial point cloud and the splats, you also have the option to visualize the Gaussians by rendering them as ellipsoids from the floating menu. SIBR has many other functionalities, please see the documentation for more details on the viewer, navigation options etc. There is also a Top View (available from the menu) that shows the placement of the input cameras and the original SfM point cloud; please note that Top View slows rendering when enabled. The real-time viewer also uses slightly more aggressive, fast culling, which can be toggled in the floating menu. If you ever encounter an issue that can be solved by turning fast culling off, please let us know.

Our COLMAP loaders expect the following dataset structure in the source path location:

<location>
|---images
|   |---<image 0>
|   |---<image 1>
|   |---...
|---sparse
    |---0
        |---cameras.bin
        |---images.bin
        |---points3D.bin

For rasterization, the camera models must be either a SIMPLE_PINHOLE or PINHOLE camera. We provide a converter script convert.py, to extract undistorted images and SfM information from input images. Optionally, you can use ImageMagick to resize the undistorted images. This rescaling is similar to MipNeRF360, i.e., it creates images with 1/2, 1/4 and 1/8 the original resolution in corresponding folders. To use them, please first install a recent version of COLMAP (ideally CUDA-powered) and ImageMagick. Put the images you want to use in a directory <location>/input.

<location>
|---input
    |---<image 0>
    |---<image 1>
    |---...

If you have COLMAP and ImageMagick on your system path, you can simply run

python convert.py -s <location> [--resize] #If not resizing, ImageMagick is not needed

Alternatively, you can use the optional parameters --colmap_executable and --magick_executable to point to the respective paths. Please note that on Windows, the executable should point to the COLMAP .bat file that takes care of setting the execution environment. Once done, <location> will contain the expected COLMAP data set structure with undistorted, resized input images, in addition to your original images and some temporary (distorted) data in the directory distorted.

If you have your own COLMAP dataset without undistortion (e.g., using OPENCV camera), you can try to just run the last part of the script: Put the images in input and the COLMAP info in a subdirectory distorted:

<location>
|---input
|   |---<image 0>
|   |---<image 1>
|   |---...
|---distorted
    |---database.db
    |---sparse
        |---0
            |---...

Then run

python convert.py -s <location> --skip_matching [--resize] #If not resizing, ImageMagick is not needed

We integrated the drop-in replacements from Taming-3dgs¹ with fused ssim into the original codebase to speed up training times. Once installed, the accelerated rasterizer delivers a $\times$ 1.6 training time speedup using --optimizer_type default and a $\times$ 2.7 training time speedup using --optimizer_type sparse_adam.

To get faster training times you must first install the accelerated rasterizer to your environment:

pip uninstall diff-gaussian-rasterization -y
cd submodules/diff-gaussian-rasterization
rm -r build
git checkout 3dgs_accel
pip install .

Then you can add the following parameter to use the sparse adam optimizer when running train.py:

--optimizer_type sparse_adam

Note that this custom rasterizer has a different behaviour than the original version, for more details on training times please see stats for training times.

1. Mallick and Goel, et al. ‘Taming 3DGS: High-Quality Radiance Fields with Limited Resources’. SIGGRAPH Asia 2024 Conference Papers, 2024, https://doi.org/10.1145/3680528.3687694, github

To have better reconstructed scenes we use depth maps as priors during optimization with each input images. It works best on untextured parts ex: roads and can remove floaters. Several papers have used similar ideas to improve various aspects of 3DGS; (e.g. DepthRegularizedGS, SparseGS, DNGaussian). The depth regularization we integrated is that used in our Hierarchical 3DGS paper, but applied to the original 3DGS; for some scenes (e.g., the DeepBlending scenes) it improves quality significantly; for others it either makes a small difference or can even be worse. For example results showing the potential benefit and statistics on quality please see here: Stats for depth regularization.

When training on a synthetic dataset, depth maps can be produced and they do not require further processing to be used in our method.

For real world datasets depth maps should be generated for each input images, to generate them please do the following:

1. Clone Depth Anything v2:

   git clone https://github.com/DepthAnything/Depth-Anything-V2.git
   

2. Download weights from Depth-Anything-V2-Large and place it under Depth-Anything-V2/checkpoints/
3. Generate depth maps:

   python Depth-Anything-V2/run.py --encoder vitl --pred-only --grayscale --img-path <path to input images> --outdir <output path>
   

4. Generate a depth_params.json file using:

   python utils/make_depth_scale.py --base_dir <path to colmap> --depths_dir <path to generated depths>
   

A new parameter should be set when training if you want to use depth regularization -d <path to depth maps>.

To compensate for exposure changes in the different input images we optimize an affine transformation for each image just as in Hierarchical 3dgs.

This can greatly improve reconstruction results for "in the wild" captures, e.g., with a smartphone when the exposure setting of the camera is not fixed. For example results showing the potential benefit and statistics on quality please see here: Stats for exposure compensation.

Add the following parameters to enable it:

--exposure_lr_init 0.001 --exposure_lr_final 0.0001 --exposure_lr_delay_steps 5000 --exposure_lr_delay_mult 0.001 --train_test_exp

Again, other excellent papers have used similar ideas e.g. NeRF-W, URF.

We added the EWA Filter from Mip Splatting in our codebase to remove aliasing. It is disabled by default but you can enable it by adding --antialiasing when training on a scene using train.py or rendering using render.py. Antialiasing can be toggled in the SIBR viewer, it is disabled by default but you should enable it when viewing a scene trained using --antialiasing. this scene was trained using --antialiasing.

Views > Top view

The Top view renders the SfM point cloud in another view with the corresponding input cameras and the Point view user camera. This allows visualization of how far the viewer is from the input cameras for example.

It is a 3D view so the user can navigate through it just as in the Point view (modes available: FPS, trackball, orbit).

Options are available to customize this view, meshes can be disabled/enabled and their scales can be modified.

A useful additional functionality is to move to the position of an input image, and progressively fade out to the SfM point view in that position (e.g., to verify camera alignment). Views from input cameras can be displayed in the Top view (note that --images-path must be set in the command line). One can snap the Top view camera to the closest input camera from the user camera in the Point view by clicking Top view settings > Cameras > Snap to closest.

OpenXR is supported in the branch gaussian_code_release_openxr Within that branch, you can find documentation for VR support here.

• Where do I get data sets, e.g., those referenced in full_eval.py? The MipNeRF360 data set is provided by the authors of the original paper on the project site. Note that two of the data sets cannot be openly shared and require you to consult the authors directly. For Tanks&Temples and Deep Blending, please use the download links provided at the top of the page. Alternatively, you may access the cloned data (status: August 2023!) from HuggingFace

• How can I use this for a much larger dataset, like a city district? The current method was not designed for these, but given enough memory, it should work out. However, the approach can struggle in multi-scale detail scenes (extreme close-ups, mixed with far-away shots). This is usually the case in, e.g., driving data sets (cars close up, buildings far away). For such scenes, you can lower the --position_lr_init, --position_lr_final and --scaling_lr (x0.3, x0.1, ...). The more extensive the scene, the lower these values should be. Below, we use default learning rates (left) and --position_lr_init 0.000016 --scaling_lr 0.001" (right).

• I'm on Windows and I can't manage to build the submodules, what do I do? Consider following the steps in the excellent video tutorial here, hopefully they should help. The order in which the steps are done is important! Alternatively, consider using the linked Colab template.

• It still doesn't work. It says something about cl.exe. What do I do? User Henry Pearce found a workaround. You can you try adding the visual studio path to your environment variables (your version number might differ); C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64 Then make sure you start a new conda prompt and cd to your repo location and try this;

conda activate gaussian_splatting
cd <dir_to_repo>/gaussian-splatting
pip install submodules\diff-gaussian-rasterization
pip install submodules\simple-knn

• I'm on macOS/Puppy Linux/Greenhat and I can't manage to build, what do I do? Sorry, we can't provide support for platforms outside of the ones we list in this README. Consider using the linked Colab template.

• I don't have 24 GB of VRAM for training, what do I do? The VRAM consumption is determined by the number of points that are being optimized, which increases over time. If you only want to train to 7k iterations, you will need significantly less. To do the full training routine and avoid running out of memory, you can increase the --densify_grad_threshold, --densification_interval or reduce the value of --densify_until_iter. Note however that this will affect the quality of the result. Also try setting --test_iterations to -1 to avoid memory spikes during testing. If --densify_grad_threshold is very high, no densification should occur and training should complete if the scene itself loads successfully.

• 24 GB of VRAM for reference quality training is still a lot! Can't we do it with less? Yes, most likely. By our calculations it should be possible with way less memory (~8GB). If we can find the time we will try to achieve this. If some PyTorch veteran out there wants to tackle this, we look forward to your pull request!

• How can I use the differentiable Gaussian rasterizer for my own project? Easy, it is included in this repo as a submodule diff-gaussian-rasterization. Feel free to check out and install the package. It's not really documented, but using it from the Python side is very straightforward (cf. gaussian_renderer/__init__.py).

• Wait, but <insert feature> isn't optimized and could be much better? There are several parts we didn't even have time to think about improving (yet). The performance you get with this prototype is probably a rather slow baseline for what is physically possible.

• Something is broken, how did this happen? We tried hard to provide a solid and comprehensible basis to make use of the paper's method. We have refactored the code quite a bit, but we have limited capacity to test all possible usage scenarios. Thus, if part of the website, the code or the performance is lacking, please create an issue. If we find the time, we will do our best to address it.