ICLR 2026 Papers

The sparse unordered structure of point clouds causes unnecessary computation and energy consumption in deep models. Conventionally, the Transformer architecture is leveraged to model global relationships in point clouds, however, its quadratic complexity restricts scalability. Although the Mamba architecture enables efficient global modeling with linear complexity, it lacks natural adaptability to unordered point clouds. Spiking Neural Network (SNN) is an energy-efficient alternative to Artificial Neural Network (ANN), offering an ultra low-power event-driven paradigm. The inherent sparsity and event-driven characteristics of SNN are highly compatible with the sparse distribution of point clouds. To balance efficiency and performance, we propose a hybrid spiking Mamba-Transformer (3DSMT) model for point cloud analysis. 3DSMT integrates a Spiking Local Offset Attention module to efficiently capture fine-grained local geometric features with a spiking Mamba block designed for unordered point clouds to achieve global feature integration with linear complexity. Experiments show that 3DSMT achieves state-of-the-art performance among SNN-based methods in shape classification, few-shot classification, and part segmentation tasks, significantly reducing computational energy consumption while also outperforming numerous ANN-based models. Our source code is in supplementary material and will be made publicly available

What happens when multiple compression methods are combined—does the order in which they are applied matter? Joint model compression has emerged as a powerful strategy to achieve higher efficiency by combining multiple methods such as pruning and quantization. A central but underexplored factor in joint model compression is the compression order, or the sequence of different methods within the compression pipeline. Most prior studies have either sidestepped the issue by assuming orthogonality between techniques, while a few have examined them only in highly constrained cases. Consequently, the broader role of compression order in shaping model performance remains poorly understood. In this paper, we address the overlooked problem of compression order and provide both theoretical and empirical analysis. We formulate the problem of optimizing the compression order and introduce the Progressive Intensity Hypothesis, which states that weaker perturbations should precede stronger ones. We provide theoretical guarantees showing that the relative benefit of one order increases with the underlying performance gap. Extensive experiments on both language and vision models validate the hypothesis, and further show its generality to broader setups such as multi-stage compression and mixed-precision quantization.

Synergistically capturing intricate local structures and global contextual dependencies has become a critical challenge in point cloud representation learning. To address this, we introduce PointLearner, a point cloud representation learning network that closely aligns with biological vision which employs an active, foveation-inspired processing strategy, thus enabling local geometric modeling and long-range dependency interactions simultaneously. Specifically, we first design a point-focused attention, which simulates foveal vision at the visual focus through a competitive normalized attention mechanism between local neighbors and spatially downsampled features. The spatially downsampled features are extracted by a pooling method based on learnable inducing points, which can flexibly adapt to the non-uniform distribution of point clouds as the number of inducing points is controlled and they interact directly with point clouds. Second, we propose a context-scan state space that mimics eye's saccade inference, which infers the overall semantic structure and spatial content in the scene through a scan path guided by the Hilbert curve for the bidirectional S6. With this focus-then-context biomimetic design, PointLearner demonstrates remarkable robustness and achieves state-of-the-art performance across multiple point cloud tasks. The code is available at https://github.com/Point-Cloud-Learning/PointLearner.

Recent advances in large language models (LLMs) have improved the accuracy of visual question answering (VQA) systems. However, directly applying LLMs to VQA still presents several challenges: (a) suboptimal performance when handling questions from specialized domains, (b) higher computational costs and slower inference speed due to large model sizes, and (c) the absence of a systematic approach to precisely quantify the uncertainty of LLM responses, raising concerns about their reliability in high-stakes tasks. To address these issues, we propose an UNcertainty-aware LLM-Integrated VQA model ($\texttt{Uni-VQA}$). This model facilitates knowledge exchange between the LLM and a calibrated task specific model (\ie \texttt{TS-VQA}), guided by reliable confidence scores, resulting in improved VQA accuracy, reliability and inference speed. Our framework strategically leverages these confidence scores to manage the interaction between the LLM and $\texttt{TS-VQA}$: the specialized questions are answered by the $\texttt{TS-VQA}$ model, while general knowledge questions are handled by the LLM. For questions requiring both specialized and general knowledge, the $\texttt{TS-VQA}$ provides candidate answers, which the LLM then combines with its internal knowledge to generate a more accurate response. Extensive experiments on VQA datasets demonstrate the theoretically justified advantages of $\texttt{Uni-VQA}$ over using the LLM or $\texttt{TS-VQA}$ alone.

Generative AI has substantially facilitated realistic image synthesizing, posing great challenges for reliable forensics. When image forensic detectors are deployed in the wild, the inputs usually undergone various distortions including compression, rescaling, and lossy transmission. Such distortions severely erode forensic traces and make a detector fail silently—returning an over-confident binary prediction while being incapable of making reliable decision, as the detector cannot explicitly perceive the degree of data distortion. This paper argues that reliable forensics must therefore move beyond "is the image real or fake?" to also ask "how trustworthy is the detector's decision on the image?" We formulate this requirement as Detector's Distortion-Aware Confidence (DAC): a sample-level confidence that a given detector could properly handle the input. Taking AI-generated image detection as an example, we empirically discover that detection accuracy drops almost monotonically with full-reference image quality scores as distortion becomes severer, while such references are in fact unavailable at test time. Guided by this observation, the Distortion-Aware Confidence Model (DACOM) is proposed as a useful assistant to the forensic detector. DACOM utilizes full-reference image quality assessment to provide oracle statistical information that labels the detectability of images for training, and integrates intermediate forensic features of the detector, no-reference image quality descriptors and distortion-type cues to estimate DAC. With the estimated confidence score, it is possible to conduct selective abstention and multi-detector routing to improve the overall accuracy of a detection system. Extensive experiments have demonstrated the effectiveness of our approach.

Fine-tuning large language models (LLMs) on a mixture of diverse datasets poses challenges due to data imbalance and heterogeneity. Existing methods often address these issues across datasets (globally) but overlook the imbalance and heterogeneity within individual datasets (locally), which limits their effectiveness. We introduce Hierarchical Balancing Optimization (HBO), a novel method that enables LLMs to autonomously adjust data allocation during fine-tuning both across datasets (globally) and within each individual dataset (locally). HBO employs a bilevel optimization strategy with two types of actors: a Global Actor, which balances data sampling across different subsets of the training mixture, and several Local Actors, which optimizes data usage within each subset based on difficulty levels. These actors are guided by reward functions derived from the LLM's training state, which measure learning progress and relative performance improvement. We evaluate HBO on three LLM backbones across nine diverse tasks in multilingual and multitask setups. Results show that HBO consistently outperforms existing baselines, achieving significant accuracy gains. Our in-depth analysis further demonstrates that both the global actor and local actors of HBO effectively adjust data usage during fine-tuning. HBO provides a comprehensive solution to the challenges of data imbalance and heterogeneity in LLM fine-tuning, enabling more effective training across diverse datasets.

Large language model (LLM)-based multi-agent systems have shown strong capabilities in tasks such as code generation and collaborative reasoning. However, the effectiveness and robustness of these systems critically depend on their communication topology, which is often fixed or statically learned, ignoring real-world dynamics such as model upgrades, API (or tool) changes, or knowledge source variability. To address this limitation, we propose CARD (Conditional Agentic Graph Designer), a conditional graph-generation framework that instantiates AMACP, a protocol for adaptive multi-agent communication. CARD explicitly incorporates dynamic environmental signals into graph construction, enabling topology adaptation at both training and runtime. Through a conditional variational graph encoder and environment-aware optimization, CARD produces communication structures that are both effective and resilient to shifts in model capability or resource availability. Empirical results on HumanEval, MATH, and MMLU demonstrate that CARD consistently outperforms static and prompt-based baselines, achieving higher accuracy and robustness across diverse conditions. The source code is available at: \url{https://anonymous.4open.science/r/agentgraph-FF9A}.

Due to the real-time rendering performance, 3D Gaussian Splatting (3DGS) has emerged as the leading method for radiance field reconstruction. However, its reliance on spherical harmonics for color encoding inherently limits its ability to separate diffuse and specular components, making it challenging to accurately represent complex reflections. To address this, we propose a novel enhanced Gaussian kernel that explicitly models specular effects through view-dependent opacity. Meanwhile, we introduce an error-driven compensation strategy to improve rendering quality in existing 3DGS scenes. Our method begins with 2D Gaussian initialization and then adaptively inserts and optimizes enhanced Gaussian kernels, ultimately producing an augmented radiance field. Experiments demonstrate that our method not only surpasses state-of-the-art NeRF methods in rendering performance but also achieves greater parameter efficiency. Code will be released.

LLM-based web agents show immense promise for information seeking, yet their effectiveness on long-horizon tasks is hindered by a fundamental trade-off in context management. Prevailing ReAct-based agents suffer from context saturation as they accumulate noisy, raw histories, while methods that fixedly summarize the full history at each step risk the irreversible loss of critical details. Addressing these, we introduce AgentFold, a novel agent paradigm inspired by the human cognitive process of retrospective consolidation. AgentFold treats its context as a dynamic cognitive workspace to be actively sculpted, rather than a passive log to be filled. At each step, it learns to execute a folding operation, which manages its historical trajectory at multiple scales: it can perform granular condensations to preserve vital, fine-grained details, or deep consolidations to abstract away entire multi-step sub-tasks. The results on prominent benchmarks are striking: our AgentFold-30B-A3B agent achieves 36.2% on BrowseComp and 47.3% on BrowseComp-ZH. Notably, this performance not only surpasses or matches open-source models of a dramatically larger scale, such as the GLM-4.5-355B-A32B and the DeepSeek-V3.1-671B-A37B, but also surpasses leading proprietary agents like OpenAI's o4-mini. Model will be open-sourced.

We present Stylos, a single-forward 3D Gaussian framework for 3D style transfer that operates on unposed content, from a single image to a multi-view collection, conditioned on a separate reference style image. Stylos synthesizes a stylized 3D Gaussian scene without per-scene optimization or precomputed poses, achieving geometry-aware, view-consistent stylization that generalizes to unseen categories, scenes, and styles. At its core, Stylos adopts a Transformer backbone with two pathways: geometry predictions retain self-attention to preserve geometric fidelity, while style is injected via global cross-attention to enforce visual consistency across views. With the addition of a voxel-based 3D style loss that aligns aggregated scene features to style statistics, Stylos enforces view-consistent stylization while preserving geometry. Experiments across multiple datasets demonstrate that Stylos delivers high-quality zero-shot stylization, highlighting the effectiveness of global style-content coupling, the proposed 3D style loss, and the scalability of our framework from single view to large-scale multi-view settings. Our codes are available at https://github.com/HanzhouLiu/Stylos.

Fast and flexible 3D scene reconstruction from unstructured image collections remains a significant challenge. We present YoNoSplat, a feedforward model that reconstructs high-quality 3D Gaussian Splatting representations from an arbitrary number of images. Our model is highly versatile, operating effectively with both posed and unposed, calibrated and uncalibrated inputs. YoNoSplat predicts local Gaussians and camera poses for each view, which are aggregated into a global representation using either predicted or provided poses. To overcome the inherent difficulty of jointly learning 3D Gaussians and camera parameters, we introduce a novel mixing training strategy. This approach mitigates the entanglement between the two tasks by initially using ground-truth poses to aggregate local Gaussians and gradually transitioning to a mix of predicted and ground-truth poses, which prevents both training instability and exposure bias. We further resolve the scale ambiguity problem by a novel pairwise camera-distance normalization scheme and by embedding camera intrinsics into the network. Moreover, YoNoSplat also predicts intrinsic parameters, making it feasible for uncalibrated inputs. YoNoSplat demonstrates exceptional efficiency, reconstructing a scene from 100 views (at 280×518 resolution) in just 2.69 seconds on an NVIDIA GH200 GPU. It achieves state-of-the-art performance on standard benchmarks in both pose-free and pose-dependent settings. The code and pretrained models will be made public.

Category-agnostic pose estimation (CAPE) aims to localize keypoints on query images from arbitrary categories, using only a few annotated support examples for guidance. Recent approaches either treat keypoints as isolated entities or rely on manually defined skeleton priors, which are costly to annotate and inherently inflexible across diverse categories. Such oversimplification limits the model’s capacity to capture instance-wise structural cues critical for accurate pixel-level localization. To overcome these limitations, we propose \textbf{GenCape}, a \textbf{Gen}erative-based framework for \textbf{CAPE} that infers keypoint relationships solely from image-based support inputs, without additional textual descriptions or predefined skeletons. Our framework consists of two principal components: an iterative Structure-aware Variational Autoencoder (i-SVAE) and a Compositional Graph Transfer (CGT) module. The former infers soft, instance-specific adjacency matrices from support features through variational inference, embedded layer-wise into the Graph Transformer Decoder for progressive structural priors refinement. The latter adaptively aggregates multiple latent graphs into a query-aware structure via Bayesian fusion and attention-based reweighting, enhancing resilience to visual uncertainty and support-induced bias. This structure-aware design facilitates effective message propagation among keypoints and promotes semantic alignment across object categories with diverse keypoint topologies. Experimental results on the MP-100 dataset show that our method achieves substantial gains over graph-support baselines under both 1- and 5-shot settings, while maintaining competitive performance against text-support counterparts.

We present Sapiens2, a model family of high-resolution transformers for human-centric vision focused on generalization, versatility, and high-fidelity outputs. Our model sizes range from 0.4 to 5 billion parameters, with native 1K resolution and hierarchical variants that support 4K. Sapiens2 substantially improves over its predecessor in both pretraining and post-training. First, to learn features that capture low-level details (for dense prediction) and high-level semantics (for zero-shot or few-label settings), we combine masked image reconstruction with self-distilled contrastive objectives. Our evaluations show that this unified pretraining objective is better suited for a wider range of downstream tasks. Second, along the data axis, we pretrain on a curated dataset of 1 billion high-quality human images and improve the quality and quantity of task annotations. Third, architecturally, we incorporate advances from frontier models that enable longer training schedules with improved stability. Our 4K models adopt windowed attention to reason over longer spatial context and are pretrained with 2K output resolution. Sapiens2 sets a new state-of-the-art and improves over the first generation on pose (+4 mAP), body-part segmentation (+22.3 mIoU), normal estimation (+29.2 rel-angular error) and extends to new tasks such as pointmap and albedo estimation.

Despite advancements in Multimodal Large Language Models (MLLMs), their proficiency in fine-grained video motion understanding remains critically limited. They often lack inter-frame differencing and tend to average or ignore subtle visual cues. Furthermore, while visual prompting has shown potential in static images, its application to videos' temporal complexities, particularly for fine-grained motion understanding, remains largely unexplored. We investigate whether inherent capability can be unlocked to boost MLLMs' motion perception and enable distinct visual signatures tailored to decouple object and camera motion cues. In this study, we introduce $\mathtt{MotionSight}$, a novel zero-shot method pioneering object-centric visual spotlight and motion blur as visual prompts to effectively improve fine-grained motion understanding without training. To convert this into valuable data assets, we curated $\mathtt{MotionVid-QA}$, the first large-scale dataset for fine-grained video motion understanding, with hierarchical annotations including SFT and preference data, $\Theta{(40K)}$ video clips and $\Theta{(87K)}$ QAs. Experiments show $\mathtt{MotionSight}$ achieves state-of-the-art open-source performance and competitiveness with commercial models. Using $\mathtt{MotionVid-QA}$, we fine-tuned $\mathtt{MotionChat}$ on Qwen2.5VL-7B, which attains 48.3\% overall accuracy on FAVOR-Bench that is comparable to Qwen2.5VL-72B's 48.1\%. In summary, we present a novel zero-shot method and a large-scale, high-quality dataset specifically for fine-grained motion understanding. All the code and annotations will be publicly available.

This work presents Prior Depth Anything, a framework that combines incomplete but precise metric information in depth measurement with relative but complete geometric structures in depth prediction, generating accurate, dense, and detailed metric depth maps for any scene. To this end, we design a coarse-to-fine pipeline to progressively integrate the two complementary depth sources. First, we introduce pixel-level metric alignment and distance-aware weighting to pre-fill diverse metric priors by explicitly using depth prediction. It effectively narrows the domain gap between prior patterns, enhancing generalization across varying scenarios. Second, we develop a conditioned monocular depth estimation (MDE) model to refine the inherent noise of depth priors. By conditioning on the normalized pre-filled prior and prediction, the model further implicitly merges the two complementary depth sources. Our model showcases impressive zero-shot generalization across depth completion, super-resolution, and inpainting over 7 real-world datasets, matching or even surpassing previous task-specific methods. More importantly, it performs well on challenging, unseen mixed priors and enables test-time improvements by switching prediction models, providing a flexible accuracy-efficiency trade-off while evolving with advancements in MDE models.

Real-World Super-Resolution (RWSR) aims to reconstruct high-resolution images from low-resolution inputs captured under complex, real-life conditions, where diverse distortions result in significant degradation heterogeneity. Many methods rely on degradation representations, yet they struggle with the lack of spatially variant degradation modeling and degradation-content entanglement. We propose Spatially Amortized Variational Learning (SAVL), an implicit framework that models per-pixel degradations as spatially varying Gaussians inferred from local neighborhoods. SAVL couples a conditional likelihood lane (SAVL-LM) with a mutual information suppression lane (SAVL-MIS) to filter out degradation-irrelevant signals, yielding a well-constrained solution space. Both our qualitative visualizations and quantitative analyses confirm that the learned representations effectively capture the spatial distribution of complex degradations while being highly discriminative of diverse underlying degradation factors. Building on these representations, we design a degradation-aware SR network with channel-wise guidance and spatial attention modulation for adaptive reconstruction under heterogeneous degradations. Extensive experiments on real-world datasets demonstrate consistent gains over prior methods.

Complex dynamic scenes present significant challenges for predicting human behavior due to the abundance of interaction information, such as human-human and human-environment interactions. These factors complicate the analysis and understanding of human behavior, thereby increasing the uncertainty in forecasting human motions. Existing motion prediction methods thus struggle in these complex scenarios. In this paper, we propose an effective method for human motion forecasting in dynamic scenes. To achieve a comprehensive representation of interactions, we design a hierarchical interaction feature representation so that high-level features capture the overall context of the interactions, while low-level features focus on fine-grained details. Besides, we propose a coarse-to-fine interaction reasoning module that leverages both spatial and frequency perspectives to efficiently utilize hierarchical features, thereby enhancing the accuracy of motion predictions. Our method achieves state-of-the-art performance across four public datasets. We will release our code upon publication.

Improving vision-language models (VLMs) in the post-training stage typically relies on supervised fine-tuning or reinforcement learning, methods that necessitate costly, human-annotated data. While self-supervised techniques such as self-consistency have proven effective for enhancing reasoning capabilities, their application to perceptual domains such as image quality assessment (IQA) remains largely unexplored. In this work, we introduce EvoQuality, a novel framework that enables a VLM to autonomously refine its quality perception capabilities without any ground-truth labels. EvoQuality adapts the principle of self-consistency to the ranking-based nature of IQA. It generates pseudo-labels by performing pairwise majority voting on the VLM's own outputs to establish a consensus on relative quality. These pseudo-rankings are then formulated into a fidelity reward that guides the model's iterative evolution through group relative policy optimization (GRPO). By iteratively leveraging its own predictions, EvoQuality progressively refines the VLM’s perceptual capability. Extensive experiments show that EvoQuality boosts the base VLM's zero-shot performance by 31.8% on PLCC across diverse IQA benchmarks. Remarkably, despite being entirely self-supervised, EvoQuality achieves performance that is competitive with, or even surpasses, state-of-the-art supervised VLM-based IQA models, outperforming these models on 5 out of 7 IQA benchmarks. Furthermore, the framework demonstrates significant flexibility, allowing it to be stacked with pre-trained IQA models to bolster generalization on unseen datasets.

We construct VisualPRM400K, a dataset comprising about 400K multimodal process supervision data. Building upon this dataset, we develop VisualPRM, an advanced multimodal Process Reward Model (PRM) capable of estimating the value score of each step during the reasoning process. Under the Best-of-N evaluation setting, our model improves the reasoning performance of three types of MLLMs and four different model scales. Even when applied to the highly capable InternVL2.5-78B, it achieves a 5.9-point improvement across seven multimodal reasoning benchmarks. Experimental results show that the PRM model trained on our VisualPRM400K exhibits superior performance compared to Outcome Reward Models and Self-Consistency during BoN evaluation. To further facilitate the development of multimodal PRMs, we construct VisualProcessBench, a benchmark designed to measure the abilities of PRMs and MLLMs to detect incorrect steps in multimodal reasoning tasks. We hope that our work can inspire more future research and contribute to the development of MLLMs. Our model, data, and benchmark will be released.