GSplatLoc: Grounding Keypoint Descriptors into 3D Gaussian Splatting for Improved Visual Localization

[Paper] [Project Page] [Video]

Authors: Gennady Sidorov, Malik Mohrat, Denis Gridusov, Ruslan Rakhimov, Sergey Kolyubin

This repository contains the code for the paper "GSplatLoc: Grounding Keypoint Descriptors into 3D Gaussian Splatting for Improved Visual Localization".

Environment setup

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

<!-- ``` conda env create --file environment.yml conda activate gsplatloc ``` -->

conda create --name gsplatloc python=3.10
conda activate gsplatloc

PyTorch (Please check your CUDA version, we used 11.8)

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Required packages

pip install -r requirements.txt

Submodules

pip install submodules/diff-gaussian-rasterization # Rasterizer for RGB, n-dim feature, depth
pip install submodules/simple-knn

Data preparation

For main evaluation, we used 7Scenes and Cambridge Landmarks.

Below are the instructions to download and prepare the datasets.

7Scenes

You can use the datasets/setup_7scenes.py script to download and prepare the data. We experimented with Pseudo Ground Truth (PGT) camera poses obtained after running SfM on the scenes, as they are more precise than the original D-SLAM poses.

To download and prepare the datasets using the PGT poses:

# Downloads the data to datasets/pgt_7scenes_{chess, fire, ...}
python datasets/setup_7scenes.py --poses pgt

To complete the dataset preparation, follow these additional steps:

1. Download the SfM models from the visloc_pseudo_gt_limitations repository.
2. Extract the downloaded models into the datasets/ folder.

These SfM model point clouds are used for initializing the 3D Gaussian Splatting (3DGS) process.

cd datasets

# Downloads sfm models for 7scenes
gdown https://drive.google.com/uc?id=1ATijcGCgK84NKB4Mho4_T-P7x8LSL80m 
unzip 7scenes_reference_models.zip && rm 7scenes_reference_models.zip

Cambridge Landmarks

You can download and prepare the Cambridge Landmarks dataset using the script:

cd datasets

# Downloads the data to datasets/Cambridge_{GreatCourt, KingsCollege, ...}
python datasets/setup_cambridge.py

Training

python train.py -s data/DATASET_NAME -m output/OUTPUT_NAME --iterations 7000

<details> <summary><span style="font-weight: bold;">Command Line Arguments for train.py</span></summary>

--source_path / -s

Path to the source directory containing a COLMAP or Synthetic NeRF data set.

--model_path / -m

Path where the trained model should be stored (output/<random> by default).

--images / -i

Alternative subdirectory for COLMAP images (images by default).

--eval

Add this flag to use a MipNeRF360-style training/test split for evaluation.

--resolution / -r

Specifies resolution of the loaded images before training. If provided 1, 2, 4 or 8, uses original, 1/2, 1/4 or 1/8 resolution, respectively. If proveided 0, use GT feature map's resolution. For all other values, rescales the width to the given number while maintaining image aspect. If proveided -2, use the customized resolution (utils/camera_utils.py L31). If not set and input image width exceeds 1.6K pixels, inputs are automatically rescaled to this target.

--speedup

Optional speed-up module for reduced feature dimention initialization.

--data_device

Specifies where to put the source image data, cuda by default, recommended to use cpu if training on large/high-resolution dataset, will reduce VRAM consumption, but slightly slow down training. Thanks to HrsPythonix.

--white_background / -w

Add this flag to use white background instead of black (default), e.g., for evaluation of NeRF Synthetic dataset.

--sh_degree

Order of spherical harmonics to be used (no larger than 3). 3 by default.

--convert_SHs_python

Flag to make pipeline compute forward and backward of SHs with PyTorch instead of ours.

--convert_cov3D_python

Flag to make pipeline compute forward and backward of the 3D covariance with PyTorch instead of ours.

--debug

Enables debug mode if you experience erros. If the rasterizer fails, a dump file is created that you may forward to us in an issue so we can take a look.

--debug_from

Debugging is slow. You may specify an iteration (starting from 0) after which the above debugging becomes active.

--iterations

Number of total iterations to train for, 30_000 by default.

--ip

IP to start GUI server on, 127.0.0.1 by default.

--port

Port to use for GUI server, 6009 by default.

--test_iterations

Space-separated iterations at which the training script computes L1 and PSNR over test set, 7000 30000 by default.

--save_iterations

Space-separated iterations at which the training script saves the Gaussian model, 7000 30000 <iterations> by default.

--checkpoint_iterations

Space-separated iterations at which to store a checkpoint for continuing later, saved in the model directory.

--start_checkpoint

Path to a saved checkpoint to continue training from.

--quiet

Flag to omit any text written to standard out pipe.

--feature_lr

Spherical harmonics features learning rate, 0.0025 by default.

--opacity_lr

Opacity learning rate, 0.05 by default.

--scaling_lr

Scaling learning rate, 0.005 by default.

--rotation_lr

Rotation learning rate, 0.001 by default.

--position_lr_max_steps

Number of steps (from 0) where position learning rate goes from initial to final. 30_000 by default.

--position_lr_init

Initial 3D position learning rate, 0.00016 by default.

--position_lr_final

Final 3D position learning rate, 0.0000016 by default.

--position_lr_delay_mult

Position learning rate multiplier (cf. Plenoxels), 0.01 by default.

--densify_from_iter

Iteration where densification starts, 500 by default.

--densify_until_iter

Iteration where densification stops, 15_000 by default.

--densify_grad_threshold

Limit that decides if points should be densified based on 2D position gradient, 0.0002 by default.

--densification_interval

How frequently to densify, 100 (every 100 iterations) by default.

--opacity_reset_interval

How frequently to reset opacity, 3_000 by default.

--lambda_dssim

Influence of SSIM on total loss from 0 to 1, 0.2 by default.

--percent_dense

Percentage of scene extent (0--1) a point must exceed to be forcibly densified, 0.01 by default.

</details>

In this work, we didn't use the feature-3dgs speed-up module. The diff-gaussian-rasterization module is designed for 64-dimensional XFeat descriptors, but it can accommodate any 64-dimensional feature vector.

If you wish to use a different feature dimension from a different encoder, you can modify the NUM_SEMANTIC_CHANNELS parameter in the config.h file within the cuda-rasterizer directory and rebuild the module.

Localization

The main localization pipeline is implemented in loc_inference.py. Here you can find the pose prior estimation and pose refinement modules.

The basic usage is as follows:

# Specify the path to the trained model
# Additional parameters can be set as needed (see below for options)
python loc_inference.py -m output/OUTPUT_NAME

The pipeline parameters is also can be adjusted.

<details> <summary><span style="font-weight: bold;">Command Line Arguments for loc_inference.py</span></summary>

--model_path / -m

Path to the trained model directory you want to create renderings for.

--top_k

Number of top reliable keypoints from XFeat.

--ransac_iters

Number of PnP-RANSAC iterations.

--warp_lr

Learning rate for pose refinement.

--warp_iters

Number of warp iterations.

</details>

We are deeply grateful to the authors and contributors of these projects for making their code available to the research community.

License

This project is licensed under the Apache License, Version 2.0 — see the LICENSE file for details.

<section class="section" id="BibTeX"> <div class="container is-max-desktop content"> <h2 class="title">BibTeX</h2> <pre><code>@inproceedings{sidorov2025gsplatloc, title={Gsplatloc: Grounding keypoint descriptors into 3d gaussian splatting for improved visual localization}, author={Sidorov, Gennady and Mohrat, Malik and Gridusov, Denis and Rakhimov, Ruslan and Kolyubin, Sergey}, booktitle={2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, pages={12601--12607}, year={2025}, organization={IEEE} }</code></pre> </div> </section>