Publication:

POTR: Post-Training 3DGS Compression

 
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.orcid0000-0001-9530-3466
cris.virtual.orcid0000-0002-9971-3128
cris.virtual.orcid#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.orcid0000-0001-5313-4158
cris.virtualsource.department932d8640-1cb6-404a-8aad-f92227775c6e
cris.virtualsource.departmentf259279b-ee14-41d8-84a9-2136c73215e2
cris.virtualsource.department68700e30-09a4-4486-899d-1b1b761ade85
cris.virtualsource.departmentd9dcf0ec-40cb-4b14-9140-ec6762bb40e4
cris.virtualsource.orcid932d8640-1cb6-404a-8aad-f92227775c6e
cris.virtualsource.orcidf259279b-ee14-41d8-84a9-2136c73215e2
cris.virtualsource.orcid68700e30-09a4-4486-899d-1b1b761ade85
cris.virtualsource.orcidd9dcf0ec-40cb-4b14-9140-ec6762bb40e4
dc.contributor.authorRamlot, Bert
dc.contributor.authorCourteaux, Martijn
dc.contributor.authorLambert, Peter
dc.contributor.authorVan Wallendael, Glenn
dc.date.accessioned2026-10-29T15:14:02Z
dc.date.available2026-10-29T15:14:02Z
dc.date.createdwos2026
dc.date.issued2026
dc.description.abstract3D Gaussian Splatting (3DGS) has recently emerged as a promising contender to Neural Radiance Fields (NeRF) in 3D scene reconstruction and real-time novel view synthesis. 3DGS outperforms NeRF in training and inference speed but has substantially higher storage requirements. To remedy this downside, we propose POTR, a post-training 3DGS codec built on two novel techniques. First, POTR introduces a novel pruning approach that uses a modified 3DGS rasterizer to efficiently calculate every splat’s individual removal effect simultaneously. This technique results in 2- 4× fewer splats than other post-training pruning techniques and as a result also significantly accelerates inference with experiments demonstrating 1.5- 2× faster inference than other compressed models. Second, we propose a novel method to recompute lighting coefficients, significantly reducing their entropy without using any form of training. Our fast and highly parallel approach especially increases AC lighting coefficient sparsity, with experiments demonstrating increases from 70% to 97%, with minimal loss in quality. Finally, we extend POTR with a simple fine-tuning scheme to further enhance pruning, inference, and rate-distortion performance. Experiments demonstrate that POTR, even without fine-tuning, consistently outperforms all other post-training compression techniques in both rate-distortion performance and inference speed.
dc.description.wosFundingTextThis work was supported in part by the Research Foundation-Flanders (FWO) under Grant 1SA0B26N; in part by the imec.prospect Project SitSens; in part by IDLab (imec-Ghent University); in part by Flanders Innovation and Entrepreneurship (VLAIO); and in part by European Union.
dc.identifier.doi10.1109/tcsvt.2026.3685779
dc.identifier.issn1051-8215
dc.identifier.urihttps://imec-publications.be/handle/20.500.12860/60663
dc.language.isoeng
dc.provenance.editstepusergreet.vanhoof@imec.be
dc.publisherIEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
dc.source.beginpage12168
dc.source.endpage12181
dc.source.issue8
dc.source.journalIEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
dc.source.numberofpages14
dc.source.volume36
dc.title

POTR: Post-Training 3DGS Compression

dc.typeJournal article
dspace.entity.typePublication
imec.internal.crawledAt2026-09-07
imec.internal.sourcecrawler
imec.internal.wosCreatedAt2026-09-07
Files
Publication available in collections: