3D-LMVIC: Learning-based Multi-View Image Compression with 3D Gaussian Geometric Priors

Yujun Huang, Bin Chen, Niu Lian, Xin Wang, Baoyi An, Tao Dai, Shu-Tao Xia
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:25141-25158, 2025.

Abstract

Existing multi-view image compression methods often rely on 2D projection-based similarities between views to estimate disparities. While effective for small disparities, such as those in stereo images, these methods struggle with the more complex disparities encountered in wide-baseline multi-camera systems, commonly found in virtual reality and autonomous driving applications. To address this limitation, we propose 3D-LMVIC, a novel learning-based multi-view image compression framework that leverages 3D Gaussian Splatting to derive geometric priors for accurate disparity estimation. Furthermore, we introduce a depth map compression model to minimize geometric redundancy across views, along with a multi-view sequence ordering strategy based on a defined distance measure between views to enhance correlations between adjacent views. Experimental results demonstrate that 3D-LMVIC achieves superior performance compared to both traditional and learning-based methods. Additionally, it significantly improves disparity estimation accuracy over existing two-view approaches.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-huang25e, title = {3{D}-{LMVIC}: Learning-based Multi-View Image Compression with 3{D} {G}aussian Geometric Priors}, author = {Huang, Yujun and Chen, Bin and Lian, Niu and Wang, Xin and An, Baoyi and Dai, Tao and Xia, Shu-Tao}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {25141--25158}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/huang25e/huang25e.pdf}, url = {https://proceedings.mlr.press/v267/huang25e.html}, abstract = {Existing multi-view image compression methods often rely on 2D projection-based similarities between views to estimate disparities. While effective for small disparities, such as those in stereo images, these methods struggle with the more complex disparities encountered in wide-baseline multi-camera systems, commonly found in virtual reality and autonomous driving applications. To address this limitation, we propose 3D-LMVIC, a novel learning-based multi-view image compression framework that leverages 3D Gaussian Splatting to derive geometric priors for accurate disparity estimation. Furthermore, we introduce a depth map compression model to minimize geometric redundancy across views, along with a multi-view sequence ordering strategy based on a defined distance measure between views to enhance correlations between adjacent views. Experimental results demonstrate that 3D-LMVIC achieves superior performance compared to both traditional and learning-based methods. Additionally, it significantly improves disparity estimation accuracy over existing two-view approaches.} }
Endnote
%0 Conference Paper %T 3D-LMVIC: Learning-based Multi-View Image Compression with 3D Gaussian Geometric Priors %A Yujun Huang %A Bin Chen %A Niu Lian %A Xin Wang %A Baoyi An %A Tao Dai %A Shu-Tao Xia %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-huang25e %I PMLR %P 25141--25158 %U https://proceedings.mlr.press/v267/huang25e.html %V 267 %X Existing multi-view image compression methods often rely on 2D projection-based similarities between views to estimate disparities. While effective for small disparities, such as those in stereo images, these methods struggle with the more complex disparities encountered in wide-baseline multi-camera systems, commonly found in virtual reality and autonomous driving applications. To address this limitation, we propose 3D-LMVIC, a novel learning-based multi-view image compression framework that leverages 3D Gaussian Splatting to derive geometric priors for accurate disparity estimation. Furthermore, we introduce a depth map compression model to minimize geometric redundancy across views, along with a multi-view sequence ordering strategy based on a defined distance measure between views to enhance correlations between adjacent views. Experimental results demonstrate that 3D-LMVIC achieves superior performance compared to both traditional and learning-based methods. Additionally, it significantly improves disparity estimation accuracy over existing two-view approaches.
APA
Huang, Y., Chen, B., Lian, N., Wang, X., An, B., Dai, T. & Xia, S.. (2025). 3D-LMVIC: Learning-based Multi-View Image Compression with 3D Gaussian Geometric Priors. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:25141-25158 Available from https://proceedings.mlr.press/v267/huang25e.html.

Related Material