NeurIPS 2025 Papers

Accurately locating key moments within long videos is crucial for solving long video understanding (LVU) tasks. However, existing benchmarks are either severely limited in terms of video length and task diversity, or they focus solely on the end-to-end LVU performance, making them inappropriate for evaluating whether key moments can be accurately accessed. To address this challenge, we propose MomentSeeker, a novel benchmark for long-video moment retrieval (LVMR), distinguished by the following features. First, it is created based on long and diverse videos, averaging over 1,200 seconds in duration, and collected from various domains, e.g., movie, anomaly, egocentric, and sports. Second, it covers a variety of real-world scenarios in three levels: global-level, event-level, and object-level, covering common tasks like action recognition, object localization, causal reasoning, etc. Third, it incorporates rich forms of queries, including text-only queries, image-conditioned queries, and video-conditioned queries. On top of MomentSeeker, we conduct comprehensive experiments for both generation-based approaches (directly using MLLMs) and retrieval-based approaches (leveraging video retrievers). Our results reveal the significant challenges in long-video moment retrieval in terms of accuracy and efficiency, despite improvements from the latest long-video MLLMs and task-specific fine-tuning. We have publicly released MomentSeeker to facilitate future research in this area.

Real-world objects frequently undergo state transformations. From an apple being cut into pieces to a butterfly emerging from its cocoon, tracking through these changes is important for understanding real-world objects and dynamics. However, existing methods often lose track of the target object after transformation, due to significant changes in object appearance. To address this limitation, we introduce the task of Track Any State: tracking objects through transformations while detecting and describing state changes, accompanied by a new benchmark dataset, VOST-TAS. To tackle this problem, we present TubeletGraph, a zero-shot system that recovers missing objects after transformation and maps out how object states are evolving over time. TubeletGraph first identifies potentially overlooked tracks, and determines whether they should be integrated based on semantic and proximity priors. Then, it reasons about the added tracks and generates a state graph describing each observed transformation. TubeletGraph achieves state-of-the-art tracking performance under transformations, while demonstrating deeper understanding of object transformations and promising capabilities in temporal grounding and semantic reasoning for complex object transformations. Code, additional results, and the benchmark dataset are available at https://tubelet-graph.github.io.

Event cameras offer microsecond-level latency and robustness to motion blur, making them ideal for understanding dynamic environments. Yet, connecting these asynchronous streams to human language remains an open challenge. We introduce Talk2Event, the first large-scale benchmark for language-driven object grounding in event-based perception. Built from real-world driving data, Talk2Event provides over 30,000 validated referring expressions, each enriched with four grounding attributes -- appearance, status, relation to viewer, and relation to other objects -- bridging spatial, temporal, and relational reasoning. To fully exploit these cues, we propose EventRefer, an attribute-aware grounding framework that dynamically fuses multi-attribute representations through a Mixture of Event-Attribute Experts (MoEE). Our method adapts to different modalities and scene dynamics, achieving consistent gains over state-of-the-art baselines in event-only, frame-only, and event-frame fusion settings. We hope our dataset and approach will establish a foundation for advancing multimodal, temporally-aware, and language-driven perception in real-world robotics and autonomy.

We present ExAct, a new video-language benchmark for expert-level understanding of skilled physical human activities. Our new benchmark contains 3,521 expert-curated video question-answer pairs spanning 11 physical activities in 6 domains: Sports, Bike Repair, Cooking, Health, Music, and Dance. ExAct requires the correct answer to be selected from five carefully designed candidate options, thus necessitating a nuanced, fine-grained, expert-level understanding of physical human skills. Evaluating the recent state-of-the-art VLMs on ExAct reveals a substantial performance gap relative to human expert performance. Specifically, the best-performing Gemini 2.5 Pro model achieves only 55.35% accuracy, well below the 82.02% attained by trained human experts. We believe that ExAct will be beneficial for developing and evaluating VLMs capable of precise understanding of human skills in various physical and procedural domains. Dataset and code are available at https://texaser.github.io/exact_project_page/.

Despite recent advances in Large Video Language Models (LVLMs), they still struggle with fine-grained temporal understanding, hallucinate, and often make simple mistakes on even simple video question-answering tasks, all of which pose significant challenges to their safe and reliable deployment in real-world applications. To address these limitations, we propose a self-alignment framework that enables LVLMs to learn from their own errors. Our proposed framework first obtains a training set of preferred and non-preferred response pairs, where non-preferred responses are generated by incorporating common error patterns that often occur due to inadequate spatio-temporal understanding, spurious correlations between co-occurring concepts, and over-reliance on linguistic cues while neglecting the vision modality, among others. To facilitate self-alignment of LVLMs with the constructed preferred and non-preferred response pairs, we introduce Refined Regularized Preference Optimization (RRPO), a novel preference optimization method that utilizes sub-sequence-level refined rewards and token-wise KL regularization to address the limitations of Direct Preference Optimization (DPO). We demonstrate that RRPO achieves more precise alignment and more stable training compared to DPO. Our experiments and analysis validate the effectiveness of our approach across diverse video tasks, including video hallucination, short- and long-video understanding, and fine-grained temporal reasoning.

Video anomaly detection (VAD) focuses on identifying anomalies in videos. Su- pervised methods demand substantial in-domain training data and fail to deliver clear explanations for anomalies. In contrast, training-free methods leverage the knowledge reserves and language interactivity of large pre-trained models to detect anomalies. However, the current fixed-length temporal window sam- pling approaches struggle to accurately capture anomalies with varying temporal spans. Therefore, we propose VADTree that utilizes a Hierarchical Granularity- aware Tree (HGTree) structure for flexible sampling in VAD. VADTree leverages the knowledge embedded in a pre-trained Generic Event Boundary Detection (GEBD) model to characterize potential anomaly event boundaries. Specifically, VADTree decomposes the video into generic event nodes based on boundary confidence, and performs adaptive coarse-fine hierarchical structuring and re- dundancy removal to construct the HGTree. Then, the multi-dimensional priors are injected into the visual language models (VLMs) to enhance the node-wise anomaly perception, and anomaly reasoning for generic event nodes is achieved via large language models (LLMs). Finally, an inter-cluster node correlation method is used to integrate the multi-granularity anomaly scores. Extensive experiments on three challenging datasets demonstrate that VADTree achieves state-of-the-art performance in training-free settings while drastically reducing the number of sampled video segments. The code will be available at https: //github.com/wenlongli10/VADTree.

Estimating motion primitives from video (e.g., optical flow and occlusion) is a critically important computer vision problem with many downstream applications, including controllable video generation and robotics. Current solutions are primarily supervised on synthetic data or require tuning of situation-specific heuristics, which inherently limits these models' capabilities in real-world contexts. A natural solution to transcend these limitations would be to deploy large-scale, self-supervised video models, which can be trained scalably on unrestricted real-world video datasets. However, despite recent progress, motion-primitive extraction from large pretrained video models remains relatively underexplored. In this work, we describe Opt-CWM, a self-supervised flow and occlusion estimation technique from a pretrained video prediction model. Opt-CWM uses ``counterfactual probes'' to extract motion information from a base video model in a zero-shot fashion. The key problem we solve is optimizing the quality of these probes, using a combination of an efficient parameterization of the space counterfactual probes, together with a novel generic sparse-prediction principle for learning the probe-generation parameters in a self-supervised fashion. Opt-CWM achieves state-of-the-art performance for motion estimation on real-world videos while requiring no labeled data.

Visual generative and understanding models typically rely on distinct tokenizers to process images, presenting a key challenge for unifying them within a single framework. Recent studies attempt to address this by connecting the training of VQVAE (for autoregressive generation) and CLIP (for understanding) to build a unified tokenizer. However, directly combining these training objectives has been observed to cause severe loss conflicts. In this paper, we show that reconstruction and semantic supervision do not inherently conflict. Instead, the underlying bottleneck stems from limited representational capacity of discrete token space. Building on these insights, we introduce UniTok, a unified tokenizer featuring a novel multi-codebook quantization mechanism that effectively scales up the vocabulary size and bottleneck dimension. In terms of final performance, UniTok sets a new record of 0.38 rFID and 78.6\% zero-shot accuracy on ImageNet. Besides, UniTok can be seamlessly integrated into MLLMs to unlock native visual generation capability, without compromising the understanding performance. Additionally, we show that UniTok favors cfg-free generation, reducing gFID from 14.6 to 2.5 on ImageNet 256$\times$256 benchmark. All codes and models have been made publicly available.

State Space models (SSMs) like PointMamba provide efficient feature extraction for point cloud self-supervised learning with linear complexity, surpassing Transformers in computational efficiency. However, existing PointMamba-based methods rely on complex token ordering and random masking, disrupting spatial continuity and local semantic correlations. We propose \textbf{ZigzagPointMamba} to address these challenges. The key to our approach is a simple zigzag scan path that globally sequences point cloud tokens, enhancing spatial continuity by preserving the proximity of spatially adjacent point tokens. Yet, random masking impairs local semantic modeling in self-supervised learning. To overcome this, we introduce a Semantic-Siamese Masking Strategy (SMS), which masks semantically similar tokens to facilitate reconstruction by integrating local features of original and similar tokens, thus overcoming dependence on isolated local features and enabling robust global semantic modeling. Our pre-training ZigzagPointMamba weights significantly boost downstream tasks, achieving a 1.59\% mIoU gain on ShapeNetPart for part segmentation, a 0.4\% higher accuracy on ModelNet40 for classification, and 0.19\%, 1.22\%, and 0.72\% higher accuracies respectively for the classification tasks on the OBJ-BG, OBJ-ONLY, and PB-T50-RS subsets of ScanObjectNN. Code is available at https://github.com/Rabbitttttt218/ZigzagPointMamba.

Motion is one of the key components in deformable 3D scenes. Generative video models allow users to animate static scenes with text prompts for novel motion, but when it comes to 4D reconstruction, such reanimations often fall apart. The generated videos often suffer from geometric artifacts, implausible motion, and occlusions, which hinder physically consistent 4D reanimation. In this work, we introduce \textbf{Restage4D}, a geometry-preserving pipeline for deformable scene reconstruction from a single edited video. Our key insight is to leverage the unedited original video as an additional source of supervision, allowing the model to propagate accurate structure into occluded and disoccluded regions. To achieve this, we propose a video-rewinding training scheme that temporally bridges the edited and original sequences via a shared motion representation. We further introduce an occlusion-aware ARAP regularization to preserve local rigidity, and a disocclusion backtracing mechanism that supplements missing geometry in the canonical space. Together, these components enable robust reconstruction even when the edited input contains hallucinated content or inconsistent motion. We validate Restage4D on DAVIS and PointOdyssey, demonstrating improved geometry consistency, motion quality, and 3D tracking performance. Our method not only preserves deformable structure under novel motion, but also automatically corrects errors introduced by generative models, bridging the gap between flexible video synthesis and physically grounded 4D reconstruction.

Synthetic video generation is progressing very rapidly. The latest models can produce very realistic high-resolution videos that are virtually indistinguishable from real ones. Although several video forensic detectors have been recently proposed, they often exhibit poor generalization, which limits their applicability in a real-world scenario. Our key insight to overcome this issue is to guide the detector towards _seeing_ _what_ _really_ _matters_. In fact, a well-designed forensic classifier should focus on identifying intrinsic low-level artifacts introduced by a generative architecture rather than relying on high-level semantic flaws that characterize a specific model. In this work, first, we study different generative architectures, searching and identifying discriminative features that are unbiased, robust to impairments, and shared across models. Then, we introduce a novel forensic-oriented data augmentation strategy based on the wavelet decomposition and replace specific frequency-related bands to drive the model to exploit more relevant forensic cues. Our novel training paradigm improves the generalizability of AI-generated video detectors, without the need for complex algorithms and large datasets that include multiple synthetic generators. To evaluate our approach, we train the detector using data from a single generative model and test it against videos produced by a wide range of other models. Despite its simplicity, our method achieves a significant accuracy improvement over state-of-the-art detectors and obtains excellent results even on very recent generative models, such as NOVA and FLUX.

Achieving fine-grained spatio-temporal understanding in videos remains a major challenge for current Video Large Multimodal Models (Video LMMs). Addressing this challenge requires mastering two core capabilities: video referring understanding, which captures the semantics of video regions, and video grounding, which segments object regions based on natural language descriptions. However, most existing approaches tackle these tasks in isolation, limiting progress toward unified, referentially grounded video interaction. We identify a key bottleneck in the lack of high-quality, unified video instruction data and a comprehensive benchmark for evaluating referentially grounded video chat. To address these challenges, we contribute in three core aspects: dataset, model, and benchmark. First, we introduce SAMA-239K, a large-scale dataset comprising 15K videos specifically curated to enable joint learning of video referring understanding, grounding, and multi-turn video chat. Second, we propose the SAMA model, which incorporates a versatile spatio-temporal context aggregator and a Segment Anything Model to jointly enhance fine-grained video comprehension and precise grounding capabilities. Finally, we establish SAMA-Bench, a meticulously designed benchmark consisting of 5,067 questions from 522 videos, to comprehensively evaluate the integrated capabilities of Video LMMs in multi-turn, spatio-temporal referring understanding and grounded dialogue. Extensive experiments and benchmarking results show that SAMA not only achieves strong performance on SAMA-Bench but also sets a new state-of-the-art on general grounding benchmarks, while maintaining highly competitive performance on standard visual understanding benchmarks.

Differential Transformer has recently gained significant attention for its impressive empirical performance, often attributed to its ability to perform noise canceled attention. However, precisely how differential attention achieves its empirical benefits remains poorly understood. Moreover, Differential Transformer architecture demands large-scale training from scratch, hindering utilization of open pretrained weights. In this work, we conduct an in-depth investigation of Differential Transformer, uncovering three key factors behind its success: (1) enhanced expressivity via negative attention, (2) reduced redundancy among attention heads, and (3) improved learning dynamics. Based on these findings, we propose DEX, a novel method to efficiently integrate the advantages of differential attention into pretrained language models. By reusing the softmax attention scores and adding a lightweight differential operation on the output value matrix, DEX effectively incorporates the key advantages of differential attention while remaining lightweight in both training and inference. Evaluations confirm that DEX substantially improves the pretrained LLMs across diverse benchmarks, achieving significant performance gains with minimal adaptation data (< 0.01%).

Long video understanding is inherently challenging for vision-language models (VLMs) because of the extensive number of frames. With each video frame typically expanding into tens or hundreds of tokens, the limited context length of large language models (LLMs) forces the VLMs to perceive the frames sparsely and lose temporal information. To address this, we explore extreme video token compression towards *one token per frame* at the final LLM layer. Our key insight is that heuristic-based compression, widely adopted by previous methods, is prone to information loss, and this necessitates supervising LLM layers into *learnable* and *progressive* modules for *token-level compression* (LP-Comp). Such compression enables our VLM to digest 2x-4x more frames with improved performance. To further increase the token efficiency, we investigate \emph{frame-level compression}, which selects the frames most relevant to the queries via the internal attention scores of the LLM layers, named *question-conditioned compression* (QC-Comp). As a notable distinction from previous studies, we mitigate the position bias of LLM attention in long contexts, *i.e.*, the over-concentration on the beginning and end of a sequence, by splitting long videos into short segments and employing local attention. Collectively, our combined *token-level* and *frame-level* leads to an e**x**treme compression model for long video understanding, named **XComp**, achieving a significantly larger compression ratio and enabling denser frame sampling. Our XComp is finetuned from VideoChat-Flash with a data-efficient *supervised compression tuning* stage that only requires 2.5\% of the supervised fine-tuning data, yet boosts the accuracy from 42.9\% to 46.2\% on LVBench and enhances multiple other long video benchmarks.

Point Cloud Transformers have become a cornerstone in 3D representation for their ability to model long-range dependencies via self-attention. However, these models tend to overemphasize salient regions while neglecting other informative regions, which limits feature diversity and compromises robustness. To address this challenge, we introduce BlindFormer, a novel contrastive attention learning framework that redefines saliency by explicitly incorporating features typically neglected by the model. The proposed Attentional Blindspot Mining (ABM) suppresses highly attended regions during training, thereby guiding the model to explore its own blind spots. This redirection of attention expands the model’s perceptual field and uncovers richer geometric cues. To consolidate these overlooked features, BlindFormer employs Blindspot-Aware Joint Optimization (BJO), a joint learning objective that integrates blindspot feature alignment with the original pretext task. BJO enhances feature discrimination while preserving performance on the primary task, leading to more robust and generalizable representations. We validate BlindFormer on several challenging benchmarks and demonstrate consistent performance gains across multiple Transformer backbones. Notably, it improves Point-MAE by +13.4\% and PointGPT-S by +6.3\% on OBJ-BG under Gaussian noise. These results highlight the importance of mitigating attentional biases in 3D representation learning, revealing BlindFormer’s superior ability to handle perturbations and improve feature discrimination.

We present a novel video generation framework that integrates 3-dimensional geometry and dynamic awareness. To achieve this, we augment 2D videos with 3D point trajectories and align them in pixel space. The resulting 3D-aware video dataset, PointVid, is then used to fine-tune a latent diffusion model, enabling it to track 2D objects with 3D Cartesian coordinates. Building on this, we regularize the shape and motion of objects in the video to eliminate undesired artifacts, e.g., non-physical deformation. Consequently, we enhance the quality of generated RGB videos and alleviate common issues like object morphing, which are prevalent in current video models due to a lack of shape awareness. With our 3D augmentation and regularization, our model is capable of handling contact-rich scenarios such as task-oriented videos, where 3D information is essential for perceiving shape and motion of interacting solids. Our method can be seamlessly integrated into existing video diffusion models to improve their visual plausibility.

Recent advances in visual tokenizers have demonstrated their effectiveness for multimodal large language models and autoregressive generative models. However, most existing visual tokenizers rely on a fixed downsampling rate at a given visual resolution, and consequently produce a constant number of visual tokens, ignoring the fact that visual information of varying complexity warrant different token budgets. Motivated by this observation, we propose an adaptive video tokenizer "VaporTok" with two core contributions:Probabilistic Taildrop: We introduce a novel taildrop mechanism that learns a truncation index sampling distribution conditioned on visual complexity of the video. During both training and inference, the decoder reconstructs videos at adaptive token lengths, allocating more tokens to complex videos and fewer to simpler ones. Parallel Sample GRPO with Vapor Reward: By leveraging the probability distribution produced by probabilistic taildrop, we reformulate the visual tokenization pipeline as a sequential decision process. To optimize this process, we propose a variant of GRPO and a composite reward encompassing token efficiency, reconstruction fidelity, and generative quality, thus enabling metrics-aware adaptive tokenization across diverse objectives. Extensive experiments on standard video generation benchmarks confirm our analysis, showing that our adaptive approach matches or outperforms fixed‐rate baselines and naive taildrop while using fewer tokens.

Classifier-free guidance (CFG) is a core technique powering state-of-the-art image generation systems, yet its underlying mechanisms remain poorly understood. In this work, we first analyze CFG in a simplified linear diffusion model, where we show its behavior closely resembles that observed in the nonlinear case. Our analysis reveals that linear CFG improves generation quality via three distinct components: (i) a mean-shift term that approximately steers samples in the direction of class means, (ii) a positive Contrastive Principal Components (CPC) term that amplifies class-specific features, and (iii) a negative CPC term that suppresses generic features prevalent in unconditional data. We then verify that these insights in real-world, nonlinear diffusion models: over a broad range of noise levels, linear CFG resembles the behavior of its nonlinear counterpart. Although the two eventually diverge at low noise levels, we discuss how the insights from the linear analysis still shed light on the CFG's mechanism within the nonlinear regime.

Video aesthetic assessment, a vital area in multimedia computing, integrates computer vision with human cognition. Its progress is limited by the lack of standardized datasets and robust models, as the temporal dynamics of video and multimodal fusion challenges hinder direct application of image-based methods. This study introduces VADB, the largest video aesthetic database with 10,490 diverse videos annotated by 37 professionals across multiple aesthetic dimensions, including overall and attribute-specific aesthetic scores, rich language comments and objective tags. We propose VADB-Net, a dual-modal pre-training framework with a two-stage training strategy, which outperforms existing video quality assessment models in scoring tasks and supports downstream video aesthetic assessment tasks. The dataset and source code are available at https://github.com/BestiVictory/VADB.