Merge branch 'main' of github.com:w-m/3dgs-compression-survey

data_extraction/latex/3dgs_compression_survey.bbl

@@ -0,0 +1,131 @@
+\begin{thebibliography}{10}
+
+\bibitem{MipNeRF360}
+J.~T. Barron, B.~Mildenhall, D.~Verbin, P.~P. Srinivasan, and P.~Hedman.
+\newblock Mip-nerf 360: Unbounded anti-aliased neural radiance fields.
+\newblock In {\em Proceedings of the IEEE/CVF Conference on Computer Vision and
+  Pattern Recognition}, pages 5470--5479, 2022.
+
+\bibitem{chen2024hac}
+Y.~Chen, Q.~Wu, J.~Cai, M.~Harandi, and W.~Lin.
+\newblock Hac: Hash-grid assisted context for 3d gaussian splatting
+  compression, 2024, 2403.14530.
+
+\bibitem{chen2024far}
+Y.~Chen, Q.~Wu, M.~Harandi, and J.~Cai.
+\newblock How far can we compress instant-ngp-based nerf?
+\newblock In {\em Proceedings of the IEEE/CVF Conference on Computer Vision and
+  Pattern Recognition}, pages 20321--20330, 2024.
+
+\bibitem{fan2024lightgaussian}
+Z.~Fan, K.~Wang, K.~Wen, Z.~Zhu, D.~Xu, and Z.~Wang.
+\newblock Lightgaussian: Unbounded 3d gaussian compression with 15x reduction
+  and 200+ fps, 2024, 2311.17245.
+
+\bibitem{fei20243d}
+B.~Fei, J.~Xu, R.~Zhang, Q.~Zhou, W.~Yang, and Y.~He.
+\newblock 3d gaussian splatting as new era: A survey.
+\newblock {\em IEEE Transactions on Visualization and Computer Graphics}, 2024.
+
+\bibitem{girish2024eagles}
+S.~Girish, K.~Gupta, and A.~Shrivastava.
+\newblock Eagles: Efficient accelerated 3d gaussians with lightweight
+  encodings, 2024, 2312.04564.
+
+\bibitem{DeepBlending}
+P.~Hedman, J.~Philip, T.~Price, J.-M. Frahm, G.~Drettakis, and G.~Brostow.
+\newblock Deep blending for free-viewpoint image-based rendering.
+\newblock {\em ACM Transactions on Graphics (ToG)}, 37(6):1--15, 2018.
+
+\bibitem{hu2024gsplat}
+J.~Hu, R.~Li, V.~Ye, and A.~Kanazawa.
+\newblock gsplat compression, 2024.
+
+\bibitem{kerbl3Dgaussians}
+B.~Kerbl, G.~Kopanas, T.~Leimk{\"u}hler, and G.~Drettakis.
+\newblock 3d gaussian splatting for real-time radiance field rendering.
+\newblock {\em ACM Transactions on Graphics}, 42(4), July 2023.
+
+\bibitem{TanksAndTemples}
+A.~Knapitsch, J.~Park, Q.-Y. Zhou, and V.~Koltun.
+\newblock Tanks and temples: Benchmarking large-scale scene reconstruction.
+\newblock {\em ACM Transactions on Graphics}, 36(4), 2017.
+
+\bibitem{lee2024compact}
+J.~C. Lee, D.~Rho, X.~Sun, J.~H. Ko, and E.~Park.
+\newblock Compact 3d gaussian representation for radiance field, 2024.
+
+\bibitem{li2023compressing}
+L.~Li, Z.~Shen, Z.~Wang, L.~Shen, and L.~Bo.
+\newblock Compressing volumetric radiance fields to 1 mb.
+\newblock In {\em Proceedings of the IEEE/CVF Conference on Computer Vision and
+  Pattern Recognition}, pages 4222--4231, 2023.
+
+\bibitem{lu2024scaffold}
+T.~Lu, M.~Yu, L.~Xu, Y.~Xiangli, L.~Wang, D.~Lin, and B.~Dai.
+\newblock Scaffold-gs: Structured 3d gaussians for view-adaptive rendering,
+  2024.
+
+\bibitem{mildenhall2020nerf}
+B.~Mildenhall, P.~P. Srinivasan, M.~Tancik, J.~T. Barron, R.~Ramamoorthi, and
+  R.~Ng.
+\newblock Nerf: Representing scenes as neural radiance fields for view
+  synthesis.
+\newblock In {\em ECCV}, 2020.
+
+\bibitem{SyntheticNeRF}
+B.~Mildenhall, P.~P. Srinivasan, M.~Tancik, J.~T. Barron, R.~Ramamoorthi, and
+  R.~Ng.
+\newblock Nerf: Representing scenes as neural radiance fields for view
+  synthesis.
+\newblock {\em Communications of the ACM}, 65(1):99--106, 2021.
+
+\bibitem{morgenstern2024compact}
+W.~Morgenstern, F.~Barthel, A.~Hilsmann, and P.~Eisert.
+\newblock Compact 3d scene representation via self-organizing gaussian grids,
+  2024, 2312.13299.
+
+\bibitem{navaneet2023compact3d}
+K.~Navaneet, K.~P. Meibodi, S.~A. Koohpayegani, and H.~Pirsiavash.
+\newblock Compact3d: Compressing gaussian splat radiance field models with
+  vector quantization, 2024, 2311.18159.
+
+\bibitem{niedermayr2024compressed}
+S.~Niedermayr, J.~Stumpfegger, and R.~Westermann.
+\newblock Compressed 3d gaussian splatting for accelerated novel view
+  synthesis, 2024.
+
+\bibitem{papantonakis2024reducing}
+P.~Papantonakis, G.~Kopanas, B.~Kerbl, A.~Lanvin, and G.~Drettakis.
+\newblock Reducing the memory footprint of 3d gaussian splatting.
+\newblock {\em Proceedings of the ACM on Computer Graphics and Interactive
+  Techniques}, 7(1):1--17, May 2024.
+
+\bibitem{sun2024f3dgs}
+X.~Sun, J.~C. Lee, D.~Rho, J.~H. Ko, U.~Ali, and E.~Park.
+\newblock F-3dgs: Factorized coordinates and representations for 3d gaussian
+  splatting, 2024, 2405.17083.
+
+\bibitem{wang2024end}
+H.~Wang, H.~Zhu, T.~He, R.~Feng, J.~Deng, J.~Bian, and Z.~Chen.
+\newblock End-to-end rate-distortion optimized 3d gaussian representation,
+  2024, 2406.01597.
+
+\bibitem{wu2024implicit}
+M.~Wu and T.~Tuytelaars.
+\newblock Implicit gaussian splatting with efficient multi-level tri-plane
+  representation.
+\newblock {\em arXiv preprint arXiv:2408.10041}, 2024.
+
+\bibitem{wu2024recent}
+T.~Wu, Y.-J. Yuan, L.-X. Zhang, J.~Yang, Y.-P. Cao, L.-Q. Yan, and L.~Gao.
+\newblock Recent advances in 3d gaussian splatting.
+\newblock {\em Computational Visual Media}, pages 1--30, 2024.
+
+\bibitem{xie2024mesongs}
+S.~Xie, W.~Zhang, C.~Tang, Y.~Bai, R.~Lu, S.~Ge, and Z.~Wang.
+\newblock Mesongs: Post-training compression of 3d gaussians via efficient
+  attribute transformation.
+\newblock In {\em European Conference on Computer Vision}. Springer, 2024.
+
+\end{thebibliography}

data_extraction/latex/3dgs_compression_survey.tex

@@ -80,7 +80,7 @@ \section{Scope of this survey}
 In this survey, we focus on compression methods for 3D Gaussian Splatting (3DGS), aiming to optimize memory usage while preserving visual quality and real-time rendering speed. We provide a comprehensive comparison of various compression techniques, with quantitative results for the most commenly used datasets summarized in a tabulated format. Our goal is to ensure transparency and reproducibility of the included approaches. Additionally, we offer a brief explanation of each pipeline and discuss main compression approaches. Rather than covering all existing 3DGS methods, our focus is specifically on their compression techniques; for a broader overview of 3DGS methods and applications, we refer readers to \cite{wu2024recent,fei20243d}. While we include many common approaches shared between neural radiance field (NeRF)\cite{mildenhall2020nerf} compression and 3DGS compression, we direct readers to \cite{li2023compressing,chen2024far} for NeRF-specific compression methods.
 
 \input{3dgs_table}
-
+\input{datasets_and_evaluation_statistics}
 \input{3dgs_survey_text}
 
 % Bibliography (if needed)

data_extraction/latex/3dgs_survey_text.tex

@@ -1,9 +1,4 @@
-% \section{Datasets and Evaluation Statistics}
-%     % describe all datasets briefly
-%     % describe eval statistics used in the table
-
 % \section{Fundamentals of 3D Gaussian Splatting and Compression}
-
 %     \subsection{3D Gaussian Splatting}
 
 %         3D Gaussian Splatting (3DGS)\cite{kerbl3Dgaussians} is a 3D scene representation based on rasterization used to perform novel view synthesis. While more traditonal methods rely on polygonal meshes or voxel grids, 3D Gaussian Splatting relys on a set of overlapping Gaussian functions (or "splats") to model the appearence of surfaces or volumes in 3D space. A "splat" in 3DGS is a 3D Gaussian ditribution that is described by its position (XYZ), Covariance (stretch and scale), color (RGB) and Alpha (transparency). 

data_extraction/latex/build_latex.py

@@ -206,6 +206,8 @@ def extract_title_and_text(markdown: str):
     # check for html
     clean = re.compile("<.*?>")
     text = re.sub(clean, "", text)
+    # excape % character
+    text = text.replace("%", r"\%")
     return title, text
 
 
@@ -288,8 +290,6 @@ def cp_images(src_folder, dst_folder):
         "3dgs_compression_survey.blg",
         "3dgs_compression_survey.log",
         "3dgs_compression_survey.out",
-        "3dgs_contributions.tex",
-        "3dgs_table.aux",
         "dataset.bib",
         "methods.bib",
     ]

data_extraction/latex/datasets_and_evaluation_statistics.tex

@@ -0,0 +1,10 @@
+\section{Datasets and Evaluation Statistics}
+
+\subsection{Datasets}
+% % describe all datasets briefly
+Performance and quality assessment of 3D Gaussian Splatting algorithms is typically performed on multiple datasets. These datasets provide 3D scenes or objects with various properties, such as varying levels of detail, lighting conditions, and complexities, which allow for comprehensive evaluation of the algorithms. \\
+In our survey we include Tanks and Temples\cite{TanksAndTemples}, MipNerf360\cite{MipNeRF360}, Deep Blending\cite{DeepBlending} as real-world datasets, and Synthetic NeRF\cite{SyntheticNeRF} as a synthetic dataset. From Tanks and Temples we include ``truck'' and ``train'' two unbounded outdoor scenes wich have a centered view point. The MipNerf360 dataset also has a centered view point but includes in- and outdoor scenes. The following scenes are included:  ``bicycle'', ``bonsai'', ``counter'', ``flowers'', ``garden'', ``kitchen'', ``room'', ``stump'', ``treehill''. From the Deep Blending dataset we include ``Dr Johnson'' and ``Palyroom'' two indoor scenes with a viewpoint directed outward. The synthetic scenes: chair, drums, ficus, hotdog, lego, material, mic, ship stem from the SyntheticNeRF dataset.
+
+\subsection{Evaluation Statistics}
+% !!! check gsplat-protocol / check downsampling factor in 3DGS; maybe in code. 
+% % describe eval statistics used in the table