NeurIPS 2025 论文列表

Recent advances in Vision Transformers (ViTs) have shown remarkable performance across vision tasks, yet their deep, uniform layer structure introduces significant computational overhead. In this work, we explore the emergent dynamics of ViT layers through the lens of energy-based memory systems, drawing a connection between self-attention and modern Hopfield networks. We introduce a novel metric—Layer Instability Index (LII)—derived from the operational softmax mode and its variability, to quantify the metastability of each Transformer layer over time. Our analysis reveals that certain layers exhibit consistent convergence to attractor-like states, suggesting functional specialisation and early stabilisation. Leveraging this insight, we propose an adaptive training framework that dynamically freezes or skips stable layers based on their energy landscape behavior. Our method reduces training costs while maintaining or improving accuracy. Extensive experiments on ViT-S/B/L on CUB-200-2011, CIFAR-10/100, Food-101, Stanford Dogs, and Beans demonstrate the generality and efficiency of our approach. This work provides new theoretical and practical perspectives for energy-aware optimisation of deep Transformer models.

Accurately tracking camera intrinsics is crucial for achieving 3D understanding from 2D video. However, most 3D algorithms assume that camera intrinsics stay constant throughout a video, which is often not true for many real-world in-the-wild videos. A major obstacle in this field is a lack of dynamic camera intrinsics benchmarks--existing benchmarks typically offer limited diversity in scene content and intrinsics variation, and none provide per-frame intrinsic changes for consecutive video frames. In this paper, we present Intrinsics in Flux (InFlux), a real-world benchmark that provides per-frame ground truth intrinsics annotations for videos with dynamic intrinsics. Compared to prior benchmarks, InFlux captures a wider range of intrinsic variations and scene diversity, featuring 143K+ annotated frames from 386 high-resolution indoor and outdoor videos with dynamic camera intrinsics. To ensure accurate per-frame intrinsics, we build a comprehensive lookup table of calibration experiments and extend the Kalibr toolbox to improve its accuracy and robustness. Using our benchmark, we evaluate existing baseline methods for predicting camera intrinsics and find that most struggle to achieve accurate predictions on videos with dynamic intrinsics. For the dataset, code, videos, and submission, please visit https://influx.cs.princeton.edu/.

4D Gaussian Splatting (4DGS) has recently gained considerable attention as a method for reconstructing dynamic scenes. Despite achieving superior quality, 4DGS typically requires substantial storage and suffers from slow rendering speed. In this work, we delve into these issues and identify two key sources of temporal redundancy. (Q1) \textbf{Short-Lifespan Gaussians}: 4DGS uses a large portion of Gaussians with short temporal span to represent scene dynamics, leading to an excessive number of Gaussians. (Q2) \textbf{Inactive Gaussians}: When rendering, only a small subset of Gaussians contributes to each frame. Despite this, all Gaussians are processed during rasterization, resulting in redundant computation overhead. To address these redundancies, we present \textbf{4DGS-1K}, which runs at over 1000 FPS on modern GPUs. For Q1, we introduce the Spatial-Temporal Variation Score, a new pruning criterion that effectively removes short-lifespan Gaussians while encouraging 4DGS to capture scene dynamics using Gaussians with longer temporal spans. For Q2, we store a mask for active Gaussians across consecutive frames, significantly reducing redundant computations. Compared to vanilla 4DGS, our method achieves a $41\times$ reduction in storage and $9\times$ faster rasterization on complex dynamic scenes, while maintaining comparable visual quality.

This paper addresses metric 3D reconstruction of indoor scenes by exploiting their inherent geometric regularities with compact representations. Using planar 3D primitives -- a well-suited representation for man-made environments -- we introduce PLANA3R, a pose-free framework for metric $\underline{Plana}$r $\underline{3}$D $\underline{R}$econstruction from unposed two-view images. Our approach employs Vision Transformers to extract a set of sparse planar primitives, estimate relative camera poses, and supervise geometry learning via planar splatting, where gradients are propagated through high-resolution rendered depth and normal maps of primitives. Unlike prior feedforward methods that require 3D plane annotations during training, PLANA3R learns planar 3D structures without explicit plane supervision, enabling scalable training on large-scale stereo datasets using only depth and normal annotations. We validate PLANA3R on multiple indoor-scene datasets with metric supervision and demonstrate strong generalization to out-of-domain indoor environments across diverse tasks under metric evaluation protocols, including 3D surface reconstruction, depth estimation, and relative pose estimation. Furthermore, by formulating with planar 3D representation, our method emerges with the ability for accurate plane segmentation. The project page is available at: \url{https://lck666666.github.io/plana3r/}.

Free-moving object reconstruction from monocular video remains challenging, particularly without reliable pose or depth cues and under arbitrary object motion. We introduce OnlineSplatter, a novel online feed-forward framework generating high-quality, object-centric 3D Gaussians directly from RGB frames without requiring camera pose, depth priors, or bundle optimization. Our approach anchors reconstruction using the first frame and progressively refines the object representation through a dense Gaussian primitive field, maintaining constant computational cost regardless of video sequence length. Our core contribution is a dual-key memory module combining latent appearance-geometry keys with explicit directional keys, robustly fusing current frame features with temporally aggregated object states. This design enables effective handling of free-moving objects via spatial-guided memory readout and an efficient sparsification mechanism, ensuring comprehensive yet compact object coverage. Evaluations on real-world datasets demonstrate that OnlineSplatter significantly outperforms state-of-the-art pose-free reconstruction baselines, consistently improving with more observations while maintaining constant memory and runtime.

Semantic Scene Completion (SSC) has emerged as a pivotal approach for jointly learning scene geometry and semantics, enabling downstream applications such as navigation in mobile robotics. The recent generalization to Panoptic Scene Completion (PSC) advances the SSC domain by integrating instance-level information, thereby enhancing object-level sensitivity in scene understanding. While PSC was introduced using LiDAR modality, methods based on camera images remain largely unexplored. Moreover, recent Transformer-based approaches utilize a fixed set of learned queries to reconstruct objects within the scene volume. Although these queries are typically updated with image context during training, they remain static at test time, limiting their ability to dynamically adapt specifically to the observed scene. To overcome these limitations, we propose IPFormer, the first method that leverages context-adaptive instance proposals at train and test time to address vision-based 3D Panoptic Scene Completion. Specifically, IPFormer adaptively initializes these queries as panoptic instance proposals derived from image context and further refines them through attention-based encoding and decoding to reason about semantic instance-voxel relationships. Extensive experimental results show that our approach achieves state-of-the-art in-domain performance, exhibits superior zero-shot generalization on out-of-domain data, and achieves a runtime reduction exceeding 14$\times$. These results highlight our introduction of context-adaptive instance proposals as a pioneering effort in addressing vision-based 3D Panoptic Scene Completion.

We introduce GS2E (Gaussian Splatting to Event Generation), a large-scale synthetic event dataset designed for high-fidelity event vision tasks, captured from real-world sparse multi-view RGB images. Existing event datasets are often synthesized from dense RGB videos, which typically suffer from limited viewpoint diversity and geometric inconsistency, or rely on expensive, hard-to-scale hardware setups. GS2E addresses these limitations by first reconstructing photorealistic static scenes using 3D Gaussian Splatting, followed by a novel, physically-informed event simulation pipeline. This pipeline integrates adaptive trajectory interpolation with physically-consistent event contrast threshold modeling. As a result, it generates temporally dense and geometrically consistent event streams under diverse motion and lighting conditions, while maintaining strong alignment with the underlying scene structure. Experimental results on event-based 3D reconstruction highlight GS2E’s superior generalization capabilities and its practical value as a benchmark for advancing event vision research.

In this paper, we introduce LangSplatV2, which achieves high-dimensional feature splatting at 476.2 FPS and 3D open-vocabulary text querying at 384.6 FPS for high-resolution images, providing a 42 × speedup and a 47 × boost over LangSplat respectively, along with improved query accuracy. LangSplat employs Gaussian Splatting to embed 2D CLIP language features into 3D, significantly enhancing speed and learning a precise 3D language field with SAM semantics. Such advancements in 3D language fields are crucial for applications that require language interaction within complex scenes. However, LangSplat does not yet achieve real- time performance (8.2 FPS), even with advanced A100 GPUs, severely limiting its broader application. In this paper, we first conduct a detailed time analysis of LangSplat, identifying the heavyweight decoder as the primary speed bottleneck. Our solution, LangSplatV2 assumes that each Gaussian acts as a sparse code within a global dictionary, leading to the learning of a 3D sparse coefficient field that entirely eliminates the need for a heavyweight decoder. By leveraging this sparsity, we further propose an efficient sparse coefficient splatting method with CUDA optimization, rendering high-dimensional feature maps at high quality while incurring only the time cost of splatting an ultra-low-dimensional feature. Our experimental results demonstrate that LangSplatV2 not only achieves better or competitive query accuracy but is also significantly faster. Codes and demos are available at our project page: https://langsplat-v2.github.io.

We introduce the Deformable Gaussian Splats Large Reconstruction Model (DGS-LRM), the first feed-forward method predicting deformable 3D Gaussian splats from a monocular posed video of any dynamic scene. Feed-forward scene reconstruction has gained significant attention for its ability to rapidly create digital replicas of real-world environments. However, most existing models are limited to static scenes and fail to reconstruct the motion of moving objects. Developing a feed-forward model for dynamic scene reconstruction poses significant challenges, including the scarcity of training data and the need for appropriate 3D representations and training paradigms. To address these challenges, we introduce several key technical contributions: an enhanced large-scale synthetic dataset with ground-truth multi-view videos and dense 3D scene flow supervision; a per-pixel deformable 3D Gaussian representation that is easy to learn, supports high-quality dynamic view synthesis, and enables long-range 3D tracking; and a large transformer network that achieves real-time, generalizable dynamic scene reconstruction. Extensive qualitative and quantitative experiments demonstrate that DGS-LRM achieves dynamic scene reconstruction quality comparable to optimization-based methods, while significantly outperforming the state-of-the-art predictive dynamic reconstruction method on real-world examples. Its predicted physically grounded 3D deformation is accurate and can be readily adapted for long-range 3D tracking tasks, achieving performance on par with state-of-the-art monocular video 3D tracking methods.

Physical environments and circumstances are fundamentally dynamic, yet current 3D datasets and evaluation benchmarks tend to concentrate on either dynamic scenarios or dynamic situations in isolation, resulting in incomplete comprehension. To overcome these constraints, we introduce Situat3DChange, an extensive dataset supporting three situation-aware change understanding tasks following the perception-action model: 121K question-answer pairs, 36K change descriptions for perception tasks, and 17K rearrangement instructions for the action task. To construct this large-scale dataset, Situat3DChange leverages 11K human observations of environmental changes to establish shared mental models and shared situational awareness for human-AI collaboration. These observations, enriched with egocentric and allocentric perspectives as well as categorical and coordinate spatial relations, are integrated using an LLM to support understanding of situated changes. To address the challenge of comparing pairs of point clouds from the same scene with minor changes, we propose SCReasoner, an efficient 3D MLLM approach that enables effective point cloud comparison with minimal parameter overhead and no additional tokens required for the language decoder. Comprehensive evaluation on Situat3DChange tasks highlights both the progress and limitations of MLLMs in dynamic scene and situation understanding. Additional experiments on data scaling and cross-domain transfer demonstrate the task-agnostic effectiveness of using Situat3DChange as a training dataset for MLLMs. The established dataset and source code are publicly available at: https://github.com/RuipingL/Situat3DChange.

Rotary Positional Encodings (RoPE) have emerged as a highly effective technique for one-dimensional sequences in Natural Language Processing spurring recent progress towards generalizing RoPE to higher-dimensional data such as images and videos. The success of RoPE has been thought to be due to its positional equivariance, i.e. its status as a \textit{relative} positional encoding. In this paper, we mathematically show RoPE to be one of the most general solutions for equivariant positional embedding in one-dimensional data. Moreover, we show Mixed RoPE to be the analogously general solution for $M$-dimensional data, if we require commutative generators -- a property necessary for RoPE's equivariance. However, we question the necessity of equivariance. We propose Spherical RoPE, a method analogous to Mixed RoPE, but with the assumption of anti-commutative generators -- relaxing the equivariant condition. Empirically, we find Spherical RoPE to have the equivalent learning behavior as its equivariant analogues. This strongly suggests that relative positional embeddings are not as important as is commonly believed. We expect this discovery to facilitate future work in positional encodings for vision that are faster and generalize better by removing the preconception that they must be relative.

Vision Transformers (ViTs) have emerged as the dominant architecture for visual processing tasks, demonstrating excellent scalability with increased training data and model size. However, recent work has identified the emergence of artifact tokens in ViTs that are incongruous with local semantics. These anomalous tokens degrade ViT performance in tasks that require fine-grained localization or structural coherence. An effective mitigation of this issue is the addition of register tokens to ViTs, which implicitly ''absorb'' the artifact term during training. Given the availability of existing large-scale pre-trained ViTs, in this paper we seek add register tokens to existing models without needing to re-train from scratch, which is infeasible considering their size. Specifically, we propose Post Hoc Registers (**PH-Reg**), an efficient self-distillation method that integrates registers into an existing ViT without requiring additional labeled data and full retraining. PH-Reg initializes both teacher and student networks from the same pre-trained ViT. The teacher remains frozen and unmodified, while the student is augmented with randomly initialized register tokens. By applying test-time augmentation to the teacher’s inputs, we generate denoised dense embeddings free of artifacts, which are then used to optimize only a small subset of unlocked student weights. We show that our approach can effectively reduce the number of artifact tokens, improving the segmentation and depth prediction of the student ViT under zero-shot and linear probing.

We introduce ROGR, a novel approach that reconstructs a relightable 3D model of an object captured from multiple views, driven by a generative relighting model that simulates the effects of placing the object under novel environment illuminations. Our method samples the appearance of the object under multiple lighting environments, creating a dataset that is used to train a lighting-conditioned Neural Radiance Field (NeRF) that outputs the object's appearance under any input environmental lighting. The lighting-conditioned NeRF uses a novel dual-branch architecture to encode the general lighting effects and specularities separately. The optimized lighting-conditioned NeRF enables efficient feed-forward relighting under arbitrary environment maps without requiring per-illumination optimization or light transport simulation. We evaluate our approach on the established TensoIR and Stanford-ORB datasets, where it improves upon the state-of-the-art on most metrics, and showcase our approach on real-world object captures.

The essence of 3D Gaussian Splatting (3DGS) training is to smartly allocate Gaussian primitives, expressing complex regions with more primitives and vice versa. Prior researches typically mark out under-reconstructed regions in a rendering-loss-driven manner. However, such a loss-driven strategy is often dominated by low-frequency regions, which leads to insufficient modeling of high-frequency details in texture-rich regions. As a result, it yields a suboptimal spatial allocation of Gaussian primitives. This inspires us to excavate the loss-agnostic visual prior in training views to identify complex regions that need more primitives to model. Based on this insight, we propose Complexity-Density Consistent Gaussian Splatting (CDC-GS), which allocates primitives based on the consistency between visual complexity of training views and the density of primitives. Specifically, primitives involved in rendering high visual complexity areas are categorized as modeling high complexity regions, where we leverage the high frequency wavelet components of training views to measure the visual complexity. And the density of a primitive is computed with the inverse of geometric mean of its distance to the neighboring primitives. Guided by the positive correlation between primitive complexity and density, we determine primitives to be densified as well as pruned. Extensive experiments demonstrate that our CDC-GS surpasses the baseline methods in rendering quality by a large margin using the same amount of Gaussians. And we provide insightful analysis to reveal that our method serves perpendicularly to rendering loss in guiding Gaussian primitive allocation.

Medical Visual Question Answering (Med-VQA) holds significant potential for clinical decision support, yet existing efforts primarily focus on 2D imaging with limited task diversity. This paper presents 3D-RAD, a large-scale dataset designed to advance 3D Med-VQA using radiology CT scans. The 3D-RAD dataset encompasses six diverse VQA tasks: anomaly detection, image observation, medical computation, existence detection, static temporal diagnosis, and longitudinal temporal diagnosis. It supports both open- and closed-ended questions while introducing complex reasoning challenges, including computational tasks and multi-stage temporal analysis, to enable comprehensive benchmarking. Extensive evaluations demonstrate that existing vision-language models (VLMs), especially medical VLMs exhibit limited generalization, particularly in multi-temporal tasks, underscoring the challenges of real-world 3D diagnostic reasoning. To drive future advancements, we release a high-quality training set 3D-RAD-T of 136,195 expert-aligned samples, showing that fine-tuning on this dataset could significantly enhance model performance. Our dataset and code, aiming to catalyze multimodal medical AI research and establish a robust foundation for 3D medical visual understanding, are publicly available.

Spike camera with high temporal resolution offers a new perspective on high-speed dynamic scene rendering. Most existing rendering methods rely on Neural Radiance Fields (NeRF) or 3D Gaussian Splatting (3DGS) for static scenes using a monocular spike camera. However, these methods struggle with dynamic motion, while a single camera suffers from limited spatial coverage, making it challenging to reconstruct fine details in high-speed scenes. To address these problems, we propose Spike4DGS, the first high-speed dynamic scene rendering framework with 4D Gaussian Splatting using spike camera arrays. Technically, we first build a multi-view spike camera array to validate our solution, then establish both synthetic and real-world multi-view spike-based reconstruction datasets. Then, we design a multi-view spike-based dense initialization module that obtains dense point clouds and camera poses from continuous spike streams. Finally, we propose a spike-pixel synergy constraint supervision to optimize Spike4DGS, incorporating both rendered image quality loss and dynamic spatiotemporal spike loss. The results show that our Spike4DGS outperforms state-of-the-art methods in terms of novel view rendering quality on both synthetic and real-world datasets. More details are available at https://github.com/Qinghongye/Spike4DGS.

We propose HoliGS, a novel deformable Gaussian splatting framework that addresses embodied view synthesis from long monocular RGB videos. Unlike prior 4D Gaussian splatting and dynamic NeRF pipelines, which struggle with training overhead in minute-long captures, our method leverages invertible Gaussian Splatting deformation networks to reconstruct large-scale, dynamic environments accurately. Specifically, we decompose each scene into a static background plus time-varying objects, each represented by learned Gaussian primitives undergoing global rigid transformations, skeleton-driven articulation, and subtle non-rigid deformations via an invertible neural flow. This hierarchical warping strategy enables robust free-viewpoint novel-view rendering from various embodied camera trajectories by attaching Gaussians to a complete canonical foreground shape (e.g., egocentric or third-person follow), which may involve substantial viewpoint changes and interactions between multiple actors. Our experiments demonstrate that HoliGS achieves superior reconstruction quality on challenging datasets while significantly reducing both training and rendering time compared to state-of-the-art monocular deformable NeRFs. These results highlight a practical and scalable solution for EVS in real-world scenarios. The source code will be released.

In this paper, we propose HiPoNet, an end-to-end differentiable neural network for regression, classification, and representation learning on high-dimensional point clouds. Our work is motivated by single-cell data which can have very high-dimensionality --exceeding the capabilities of existing methods for point clouds which are mostly tailored for 3D data. Moreover, modern single-cell and spatial experiments now yield entire cohorts of datasets (i.e., one data set for every patient), necessitating models that can process large, high-dimensional point-clouds at scale. Most current approaches build a single nearest-neighbor graph, discarding important geometric and topological information. In contrast, HiPoNet models the point-cloud as a set of higher-order simplicial complexes, with each particular complex being created using a reweighting of features. This method thus generates multiple constructs corresponding to different views of high-dimensional data, which in biology offers the possibility of disentangling distinct cellular processes. It then employs simplicial wavelet transforms to extract multiscale features, capturing both local and global topology from each view. We show that geometric and topological information is preserved in this framework both theoretically and empirically. We showcase the utility of HiPoNet on point-cloud level tasks, involving classification and regression of entire point-clouds in data cohorts. Experimentally, we find that HiPoNet outperforms other point-cloud and graph-based models on single-cell data. We also apply HiPoNet to spatial transcriptomics datasets using spatial coordinates as one of the views. Overall, HiPoNet offers a robust and scalable solution for high-dimensional data analysis.

Generating flexible-view 3D scenes, including 360° rotation and zooming, from single images is challenging due to a lack of 3D data. To this end, we introduce FlexWorld, a novel framework that progressively constructs a persistent 3D Gaussian splatting representation by synthesizing and integrating new 3D content. To handle novel view synthesis under large camera variations, we leverage an advanced pre-trained video model fine-tuned on accurate depth-estimated training pairs. By combining geometry-aware scene integration and optimization, FlexWorld refines the scene representation, producing visually consistent 3D scenes with flexible viewpoints. Extensive experiments demonstrate the effectiveness of FlexWorld in generating high-quality novel view videos and flexible-view 3D scenes from single images, achieving superior visual quality under multiple popular metrics and datasets compared to existing state-of-the-art methods. Additionally, FlexWorld supports extrapolating from existing 3D scenes, further extending its applicability. Qualitatively, we highlight that FlexWorld can generate high-fidelity scenes that enable 360° rotations and zooming exploration. Our code is available at https://github.com/ML-GSAI/FlexWorld.