feed-forward

LocusGS improves feed-forward 3D Gaussian Splatting by augmenting Gaussian queries with 3D anchor states (center and radius), enhancing spatial coherence and rendering quality in 3D scene reconstruction.

InfiniSplat presents a feed-forward single-image 3D Gaussian Splatting framework that uses geometry-guided sampling and query-conditioned implicit decoding to achieve surface-aligned Gaussian representation, improving large-baseline monocular view synthesis and generalizing from synthetic indoor training to open-world scenes.

This paper presents a controlled study comparing attention-only transformers (Simple Attention Networks, SANs) against standard transformers matched for parameters, compute, and depth. It finds that removing feed-forward layers largely closes the performance gap when the freed capacity is reallocated to attention depth, with the remaining deficit attributed to parametric recall.

MAGiSt3R is a multi-agent feed-forward framework that achieves real-time 3D reconstruction from monocular RGB videos at 10 FPS, using a merging model (MAGMA) to combine local point maps and pose graph optimization to reduce drift.

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.

Wiola is a novel Small Language Model architecture introducing five independently designed components—SRPE, GCLA, ATM, DSFF, and WiolaRMSNorm—aimed at improving efficiency and coherence, released in sizes from 120M to 1.5B parameters and integrated with HuggingFace Transformers.

This paper introduces a parameter-neutral replacement for transformer feed-forward layers using explicit fuzzy set operations and quantifiers over sequences. The approach achieves comparable perplexity to GELU baselines while enabling interpretable grammatical-licensing detectors, though full Boolean FFNs remain unstable.

This paper proposes a feed-forward framework that decomposes 3D scenes into instance-structured token groups from unposed multi-view images, enabling direct object-level reconstruction, segmentation, and manipulation without 3D annotations.

SpatialAvatar-0 introduces a multi-stage reconstruction method for high-quality 4D head avatars using a shared FLAME-mesh-bound Gaussian representation, achieving superior performance across benchmarks with reduced iterations.

Surflo is a feed-forward 3D reconstruction model that compresses unposed RGB views into latent tokens and decodes consistent 3D surface points via flow matching, enabling variable-resolution output and outperforming existing methods in speed.

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.

TriSplat is a feed-forward 3D reconstruction network that uses oriented triangle primitives to directly generate simulation-ready meshes from single images, bypassing expensive post-processing steps. It achieves geometry-faithful reconstructions while maintaining competitive novel-view rendering quality.

UniT is a unified feed-forward model for geometry perception using a Group Autoregressive Transformer that integrates multiple paradigms (online/offline, multi-modal, long-horizon) while maintaining metric-scale accuracy via scale-adaptive loss and queue-style KV caching. It achieves state-of-the-art performance on ten benchmarks spanning seven tasks.

FFAvatar proposes a feed-forward framework for reconstructing high-quality, animatable 3D Gaussian head avatars from few unposed images in seconds, achieving a 5.5 PSNR improvement over state-of-the-art on the NeRSemble benchmark.

VGGT-Edit proposes a feed-forward framework for text-conditioned native 3D scene editing using depth-synchronized text injection and residual field prediction, achieving superior quality and efficiency over 2D-lifting approaches.

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.