ATSplat: Compact Feed-forward 3D Gaussian Splatting with Adaptive Token Expansion

Hugging Face Daily Papers Papers
3d-gaussian-splatting feed-forward novel-view-synthesis adaptive-tokens scene-reconstruction compact-representation

Summary

ATSplat introduces a feed-forward 3D Gaussian Splatting framework that uses adaptive 3D tokens to allocate primitives based on scene complexity, achieving state-of-the-art rendering quality while reducing the number of Gaussians by over 5.7 times compared to dense methods.

3D Gaussian Splatting (3DGS) achieves high-quality novel-view synthesis by optimizing freely placed primitives in 3D and adaptively densifying them in under-reconstructed regions. However, this scene-adaptive capacity allocation is largely lost in existing feed-forward 3DGS methods, which commonly regress Gaussians at input pixels and lift them along camera rays. Such pixel-aligned formulations make the number and placement of primitives depend on image resolution and input viewpoints rather than scene complexity, resulting in dense and often redundant Gaussian sets. We present ATSplat, a feed-forward 3DGS framework that restores the adaptive allocation capability of 3DGS optimization through Adaptive 3D Tokens. ATSplat first lifts coarse patch-level depth and camera cues into sparse 3D anchor tokens, forming a compact scaffold of the scene. Each token is then regressed into local Gaussians with learnable 3D offsets, decoupling primitive placement from input image grids. An Adaptive Token Expansion module predicts a token-level uncertainty score, supervised by rendering error maps, and selectively expands high-uncertainty tokens through learnable expansion layers. This sparse-to-adaptive formulation enables ATSplat to concentrate primitives in challenging regions while maintaining a compact representation. Experiments on two representative datasets, RealEstate10K and DL3DV, show that ATSplat achieves state-of-the-art rendering quality while reducing the number of Gaussians by more than 5.7times compared with dense feed-forward 3DGS methods. From 12 input images at 512 times 960 resolution, ATSplat completes reconstruction in less than a second using a single commercial GPU, and renders high-quality novel views at 1136 FPS (512 times 960) with only 311K Gaussians.
Original Article
View Cached Full Text

Cached at: 07/24/26, 05:09 AM

Paper page - ATSplat: Compact Feed-forward 3D Gaussian Splatting with Adaptive Token Expansion

Source: https://huggingface.co/papers/2607.20417

Abstract

3DGaussianSplatting(3DGS)achieveshigh-qualitynovel-viewsynthesisbyoptimizingfreelyplacedprimitivesin3Dandadaptivelydensifyingtheminunder-reconstructedregions.However,thisscene-adaptivecapacityallocationislargelylostinexistingfeed-forward3DGSmethods,whichcommonlyregressGaussiansatinputpixelsandliftthemalongcamerarays.Suchpixel-alignedformulationsmakethenumberandplacementofprimitivesdependonimageresolutionandinputviewpointsratherthanscenecomplexity,resultingindenseandoftenredundantGaussiansets.WepresentATSplat,afeed-forward3DGSframeworkthatrestorestheadaptiveallocationcapabilityof3DGSoptimizationthroughAdaptive3DTokens.ATSplatfirstliftscoarsepatch-leveldepthandcameracuesintosparse3Danchortokens,formingacompactscaffoldofthescene.EachtokenisthenregressedintolocalGaussianswithlearnable3Doffsets,decouplingprimitiveplacementfrominputimagegrids.AnAdaptiveTokenExpansionmodulepredictsatoken-leveluncertaintyscore,supervisedbyrenderingerrormaps,andselectivelyexpandshigh-uncertaintytokensthroughlearnableexpansionlayers.Thissparse-to-adaptiveformulationenablesATSplattoconcentrateprimitivesinchallengingregionswhilemaintainingacompactrepresentation.Experimentsontworepresentativedatasets,RealEstate10KandDL3DV,showthatATSplatachievesstate-of-the-artrenderingqualitywhilereducingthenumberofGaussiansbymorethan5.7timescomparedwithdensefeed-forward3DGSmethods.From12inputimagesat512times960resolution,ATSplatcompletesreconstructioninlessthanasecondusingasinglecommercialGPU,andrendershigh-qualitynovelviewsat1136FPS(512times960)withonly311KGaussians.

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2607\.20417

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.20417 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.20417 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.20417 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens

Hugging Face Daily Papers

GlobalSplat introduces an efficient feed-forward framework for 3D Gaussian splatting that achieves compact and consistent scene reconstruction using global scene tokens, reducing computational overhead and inference time to under 78ms. The method uses a coarse-to-fine training approach to prevent representation bloat while maintaining competitive novel-view synthesis performance with significantly fewer Gaussians (16K) compared to dense baselines.

ZipSplat: Fewer Gaussians, Better Splats

Hugging Face Daily Papers

ZipSplat is a token-based feed-forward 3D Gaussian Splatting model that uses k-means clustering to decouple Gaussian placement from the pixel grid, achieving ~6x fewer Gaussians while setting new state-of-the-art results on DL3DV and RealEstate10K without requiring ground-truth poses or intrinsics.

Gaussian Point Splatting

Hacker News Top

Researchers propose Gaussian Point Splatting, a stochastic rendering method using pixel-sized opaque points and 64-bit GPU atomics that scales to hundreds of millions of Gaussians in real time. The method, accepted at SIGGRAPH 2026, employs hierarchical culling and parallel programming primitives to achieve even workload distribution with only minor noise differences compared to original Gaussian splatting.

SplatWeaver: Learning to Allocate Gaussian Primitives for Generalizable Novel View Synthesis

Hugging Face Daily Papers

SplatWeaver is a feed-forward novel view synthesis framework that dynamically allocates 3D Gaussian primitives based on spatial complexity, improving rendering quality and efficiency over fixed-allocation methods. It leverages cardinality Gaussian experts and a pixel-level routing scheme guided by high-frequency priors to adaptively distribute primitives across complex and smooth scene regions.