NeurIPS 2025 Papers

This paper presents a pioneering exploration of reinforcement learning (RL) via group relative policy optimization for unified multimodal large language models (ULMs), aimed at simultaneously reinforcing generation and understanding capabilities. Through systematic pilot studies, we uncover the significant potential of ULMs to enable the synergistic co-evolution of dual capabilities within a shared policy optimization framework. Building on this insight, we introduce \textbf{CoRL}, a \textbf{Co}-\textbf{R}einforcement \textbf{L}earning framework comprising a unified RL stage for joint optimization and a refined RL stage for task-specific enhancement. With the proposed CoRL, our resulting model, \textbf{ULM-R1}, achieves average improvements of 7\% on three text-to-image generation datasets and 23\% on nine multimodal understanding benchmarks. These results demonstrate the effectiveness of CoRL and highlight the substantial benefits of reinforcement learning in facilitating cross-task synergy and optimization for ULMs. Code is available at \url{https://github.com/mm-vl/ULM-R1}.

Dexterous manipulation is critical for advancing robot capabilities in real-world applications, yet diverse and high-quality datasets remain scarce. Existing data collection methods either rely on human teleoperation or require significant human engineering, or generate data with limited diversity, which restricts their scalability and generalization. In this paper, we introduce DexFlyWheel, a scalable data generation framework that employs a self-improving cycle to continuously enrich data diversity. Starting from efficient seed demonstrations warmup, DexFlyWheel expands the dataset through iterative cycles. Each cycle follows a closed-loop pipeline that integrates Imitation Learning (IL), residual Reinforcement Learning (RL), rollout trajectory collection, and data augmentation. Specifically, IL extracts human-like behaviors from demonstrations, and residual RL enhances policy generalization. The learned policy is then used to generate trajectories in simulation, which are further augmented across diverse environments and spatial configurations before being fed back into the next cycle. Over successive iterations, a self-improving data flywheel effect emerges, producing datasets that cover diverse scenarios and thereby scaling policy performance. Experimental results demonstrate that DexFlyWheel generates over 2,000 diverse demonstrations across four challenging tasks. Policies trained on our dataset achieve an average success rate of 81.9\% on the challenge test sets and successfully transfer to the real world through digital twin, achieving a 78.3\% success rate on dual-arm lift tasks.

Automatic indoor layout generation has attracted increasing attention due to its potential in interior design, virtual environment construction, and embodied AI. Existing methods fall into two categories: prompt-driven approaches that leverage proprietary LLM services (e.g., GPT APIs), and learning-based methods trained on layout data upon diffusion-based models. Prompt-driven methods often suffer from spatial inconsistency and high computational costs, while learning-based methods are typically constrained by coarse relational graphs and limited datasets, restricting their generalization to diverse room categories. In this paper, we revisit LLM-based indoor layout generation and present 3D-SynthPlace, a large-scale dataset that combines synthetic layouts generated via a `GPT synthesize, Human inspect' pipeline, upgraded from the 3D-Front dataset. 3D-SynthPlace contains nearly 17,000 scenes, covering four common room types—bedroom, living room, kitchen, and bathroom—enriched with diverse objects and high-level spatial annotations. We further introduce OptiScene, a strong open-source LLM optimized for indoor layout generation, fine-tuned based on our 3D-SynthPlace dataset through our two-stage training. For the warum-up stage I, we adopt supervised fine-tuning (SFT), which is taught to first generate high-level spatial descriptions then conditionally predict concrete object placements. For the reinforcing stage II, to better align the generated layouts with human design preferences, we apply multi-turn direct preference optimization (DPO), which significantly improving layout quality and generation success rates. Extensive experiments demonstrate that OptiScene outperforms traditional prompt-driven and learning-based baselines. Moreover, OptiScene shows promising potential in interactive tasks such as scene editing and robot navigation, highlighting its applicability beyond static layout generation.

Worldwide image geolocalization—the task of predicting GPS coordinates from images taken anywhere on Earth—poses a fundamental challenge due to the vast diversity in visual content across regions. While recent approaches adopt a two-stage pipeline of retrieving candidates and selecting the best match, they typically rely on simplistic similarity heuristics and point-wise supervision, failing to model spatial relationships among candidates. In this paper, we propose **GeoRanker**, a distance-aware ranking framework that leverages large vision-language models to jointly encode query–candidate interactions and predict geographic proximity. In addition, we introduce a *multi-order distance loss* that ranks both absolute and relative distances, enabling the model to reason over structured spatial relationships. To support this, we curate GeoRanking, the first dataset explicitly designed for geographic ranking tasks with multimodal candidate information. GeoRanker achieves state-of-the-art results on two well-established benchmarks (IM2GPS3K and YFCC4K), significantly outperforming current best methods. We also release our code, checkpoint, and dataset online for ease of reproduction.

Diffusion models have achieved significant progress in image generation and the pre-trained Stable Diffusion (SD) models are helpful for image deblurring by providing clear image priors. However, directly using a blurry image or a pre-deblurred one as a conditional control for SD will either hinder accurate structure extraction or make the results overly dependent on the deblurring network. In this work, we propose a Latent Kernel Prediction Network (LKPN) to achieve robust real-world image deblurring. Specifically, we co-train the LKPN in the latent space with conditional diffusion. The LKPN learns a spatially variant kernel to guide the restoration of sharp images in the latent space. By applying element-wise adaptive convolution (EAC), the learned kernel is utilized to adaptively process the blurry feature, effectively preserving the information of the blurry input. This process thereby more effectively guides the generative process of SD, enhancing both the deblurring efficacy and the quality of detail reconstruction. Moreover, the results at each diffusion step are utilized to iteratively estimate the kernels in LKPN to better restore the sharp latent by EAC in the subsequent step. This iterative refinement enhances the accuracy and robustness of the deblurring process. Extensive experimental results demonstrate that the proposed method outperforms state-of-the-art image deblurring methods on both benchmark and real-world images.

Generating high-resolution 3D shapes using volumetric representations such as Signed Distance Functions (SDFs) presents substantial computational and memory challenges. We introduce Direct3D-S2, a scalable 3D generation framework based on sparse volumes that achieves superior output quality with dramatically reduced training costs. Our key innovation is the Spatial Sparse Attention (SSA) mechanism, which greatly enhances the efficiency of Diffusion Transformer (DiT) computations on sparse volumetric data. SSA allows the model to effectively process large token sets within sparse volumes, significantly reducing computational overhead and achieving a 3.9$\times$ speedup in the forward pass and a 9.6$\times$ speedup in the backward pass. Our framework also includes a variational autoencoder (VAE) that maintains a consistent sparse volumetric format across input, latent, and output stages. Compared to previous methods with heterogeneous representations in 3D VAE, this unified design significantly improves training efficiency and stability. Our model is trained on public datasets, and experiments demonstrate that Direct3D-S2 not only surpasses state-of-the-art methods in generation quality and efficiency, but also enables training at 1024³ resolution using only 8 GPUs—a task typically requiring at least 32 GPUs for volumetric representations at $256^3$ resolution, thus making gigascale 3D generation both practical and accessible. Project page: https://www.neural4d.com/research-page/direct3d-s2.

Deciphering how visual stimuli are transformed into cortical responses is a fundamental challenge in computational neuroscience. This visual-to-neural mapping is inherently a one-to-many relationship, as identical visual inputs reliably evoke variable hemodynamic responses across trials, contexts, and subjects. However, existing deterministic methods struggle to simultaneously model this biological variability while capturing the underlying functional consistency that encodes stimulus information. To address these limitations, we propose SynBrain, a generative framework that simulates the transformation from visual semantics to neural responses in a probabilistic and biologically interpretable manner. SynBrain introduces two key components: (i) BrainVAE models neural representations as continuous probability distributions via probabilistic learning while maintaining functional consistency through visual semantic constraints; (ii) A Semantic-to-Neural Mapper acts as a semantic transmission pathway, projecting visual semantics into the neural response manifold to facilitate high-fidelity fMRI synthesis. Experimental results demonstrate that SynBrain surpasses state-of-the-art methods in subject-specific visual-to-fMRI encoding performance. Furthermore, SynBrain adapts efficiently to new subjects with few-shot data and synthesizes high-quality fMRI signals that are effective in improving data-limited fMRI-to-image decoding performance. Beyond that, SynBrain reveals functional consistency across trials and subjects, with synthesized signals capturing interpretable patterns shaped by biological neural variability. Our code is available at https://github.com/MichaelMaiii/SynBrain.

Flicker artifacts in short-exposure images are caused by the interplay between the row-wise exposure mechanism of rolling shutter cameras and the temporal intensity variations of alternating current (AC)-powered lighting. These artifacts typically appear as uneven brightness distribution across the image, forming noticeable dark bands. Beyond compromising image quality, this structured noise also affects high-level tasks, such as object detection and tracking, where reliable lighting is crucial. Despite the prevalence of flicker, the lack of a large-scale, realistic dataset has been a significant barrier to advancing research in flicker removal. To address this issue, we present BurstDeflicker, a scalable benchmark constructed using three complementary data acquisition strategies. First, we develop a Retinex-based synthesis pipeline that redefines the goal of flicker removal and enables controllable manipulation of key flicker-related attributes (e.g., intensity, area, and frequency), thereby facilitating the generation of diverse flicker patterns. Second, we capture 4,000 real-world flicker images from different scenes, which help the model better understand the spatial and temporal characteristics of real flicker artifacts and generalize more effectively to wild scenarios. Finally, due to the non-repeatable nature of dynamic scenes, we propose a green-screen method to incorporate motion into image pairs while preserving real flicker degradation. Comprehensive experiments demonstrate the effectiveness of our dataset and its potential to advance research in flicker removal.

Generative modeling of discrete variables is challenging yet crucial for applications in natural language processing and biological sequence design. We introduce the Shortlisting Model (SLM), a novel simplex-based diffusion model inspired by progressive candidate pruning. SLM operates on simplex centroids, reducing generation complexity and enhancing scalability. Additionally, SLM incorporates a flexible implementation of classifier-free guidance, enhancing unconditional generation performance. Extensive experiments on DNA promoter and enhancer design, protein design, character-level and large-vocabulary language modeling demonstrate the competitive performance and strong potential of SLM. Our code can be found at https://github.com/GenSI-THUAIR/SLM.

Recent studies have demonstrated the importance of high-quality visual representations in image generation and have highlighted the limitations of generative models in image understanding. As a generative paradigm originally designed for natural language, autoregressive models face similar challenges. In this work, we present the first systematic investigation into the mechanisms of applying the next-token prediction paradigm to the visual domain. We identify three key properties that hinder the learning of high-level visual semantics: local and conditional dependence, inter-step semantic inconsistency, and spatial invariance deficiency. We show that these issues can be effectively addressed by introducing self-supervised objectives during training, leading to a novel training framework, Self-guided Training for AutoRegressive models (ST-AR). Without relying on pre-trained representation models, ST-AR significantly enhances the image understanding ability of autoregressive models and leads to improved generation quality. Specifically, ST-AR brings approximately 42% FID improvement for LlamaGen-L and 49% FID improvement for LlamaGen-XL, while maintaining the same sampling strategy.

Recent text-to-image models have revolutionized image generation, but they still struggle with maintaining concept consistency across generated images. While existing works focus on character consistency, they often overlook the crucial role of scenes in storytelling, which restricts their creativity in practice. This paper introduces scene-oriented story generation, addressing two key challenges: (i) scene planning, where current methods fail to ensure scene-level narrative coherence by relying solely on text descriptions, and (ii) scene consistency, which remains largely unexplored in terms of maintaining scene consistency across multiple stories. We propose SceneDecorator, a training-free framework that employs VLM-Guided Scene Planning to ensure narrative coherence across different scenes in a ``global-to-local'' manner, and Long-Term Scene-Sharing Attention to maintain long-term scene consistency and subject diversity across generated stories. Extensive experiments demonstrate the superior performance of SceneDecorator, highlighting its potential to unleash creativity in the fields of arts, films, and games.

Recent interest in building foundation models for knowledge graphs has highlighted a fundamental challenge: knowledge graph data is scarce. The best-known knowl- edge graphs are primarily human-labeled, created by pattern-matching, or extracted using early NLP techniques. While human-generated knowledge graphs are in short supply, automatically extracted ones are of questionable quality. We present KGGen, a novel text-to-knowledge-graph generator that uses language models to extract high-quality graphs from plain text with a novel entity resolution approach that clusters related entities, significantly reducing the sparsity problem that plagues existing extractors. Unlike other KG generators, KGGen clusters and de-duplicates related entities to reduce sparsity in extracted KGs. Along with KGGen, we release Measure of Information in Nodes and Edges (MINE), the first benchmark to test an extractor’s ability to produce a useful KG from plain text. We benchmark our new tool against leading existing generators such as Microsoft’s GraphRAG; we achieve comparable retrieval accuracy on the generated graphs and better information re- tention. Moreover, our graphs exhibit more concise and generalizable entities and relations. Our code is open-sourced at https://github.com/stair-lab/kg-gen/.

Text-to-image diffusion models are typically trained on large-scale web data, often resulting in outputs that misalign with human preferences. Inspired by preference learning in large language models, we propose ABC (Alignment by Classification), a simple yet effective framework for aligning diffusion models with human preferences. In contrast to prior DPO-based methods that depend on suboptimal supervised fine-tuned (SFT) reference models, ABC assumes access to an ideal reference model perfectly aligned with human intent and reformulates alignment as a classification problem. Under this view, we recognize that preference data naturally forms a semi-supervised classification setting. To address this, we propose a data augmentation strategy that transforms preference comparisons into fully supervised training signals. We then introduce a classification-based ABC loss to guide alignment. Our alignment by classification approach could effectively steer the diffusion model toward the behavior of the ideal reference. Experiments on various diffusion models show that our ABC consistently outperforms existing baselines, offering a scalable and robust solution for preference-based text-to-image fine-tuning.

Controlling the spatial and semantic structure of diffusion-generated images remains a challenge. Existing methods like ControlNet rely on handcrafted condition maps and retraining, limiting flexibility and generalization. Inversion-based approaches offer stronger alignment but incur high inference cost due to dual-path denoising. We present \textbf{FreeControl}, a training-free framework for semantic structural control in diffusion models. Unlike prior methods that extract attention across multiple timesteps, FreeControl performs \textit{one-step attention extraction} from a single, optimally chosen timestep and reuses it throughout denoising. This enables efficient structural guidance without inversion or retraining. To further improve quality and stability, we introduce \textit{Latent-Condition Decoupling (LCD)}: a principled separation of the timestep condition and the noised latent used in attention extraction. LCD provides finer control over attention quality and eliminates structural artifacts. FreeControl also supports compositional control via reference images assembled from multiple sources, enabling intuitive scene layout design and stronger prompt alignment. FreeControl introduces a new paradigm for test-time control—enabling structurally and semantically aligned, visually coherent generation directly from raw images, with the flexibility for intuitive compositional design and compatibility with modern diffusion models at ~5\% additional cost.

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.

Diffusion models have achieved impressive success in high-fidelity image generation but suffer from slow sampling due to their inherently iterative denoising process. While recent one-step methods accelerate inference by learning direct noise-to-image mappings, they sacrifice the interpretability and fine-grained control intrinsic to diffusion dynamics, key advantages that enable applications like editable generation. To resolve this dichotomy, we introduce **Hierarchical Koopman Diffusion**, a novel framework that achieves both one-step sampling and interpretable generative trajectories. Grounded in Koopman operator theory, our method lifts the nonlinear diffusion dynamics into a latent space where evolution is governed by globally linear operators, enabling closed-form trajectory solutions. This formulation not only eliminates iterative sampling but also provides full access to intermediate states, allowing manual intervention during generation. To model the multi-scale nature of images, we design a hierarchical architecture that disentangles generative dynamics across spatial resolutions via scale-specific Koopman subspaces, capturing coarse-to-fine details systematically. We empirically show that the Hierarchical Koopman Diffusion not only achieves competitive one-step generation performance but also provides a principled mechanism for interpreting and manipulating the generative process through spectral analysis. Our framework bridges the gap between fast sampling and interpretability in diffusion models, paving the way for explainable image synthesis in generative modeling.

Gene expression estimation from pathology images has the potential to reduce the RNA sequencing cost. Point-wise loss functions have been widely used to minimize the discrepancy between predicted and absolute gene expression values. However, due to the complexity of the sequencing techniques and intrinsic variability across cells, the observed gene expression contains stochastic noise and batch effects, and estimating the absolute expression values accurately remains a significant challenge. To mitigate this, we propose a novel objective of learning relative expression patterns rather than absolute levels. We assume that the relative expression levels of genes exhibit consistent patterns across independent experiments, even when absolute expression values are affected by batch effects and stochastic noise in tissue samples. Based on the assumption, we model the relation and propose a novel loss function called STRank that is robust to noise and batch effects. Experiments using synthetic datasets and real datasets demonstrate the effectiveness of the proposed method. The code is available at https://github.com/naivete5656/STRank.

Conditional image generative models hold considerable promise to produce infinite amounts of synthetic training data. Yet, recent progress in generation quality has come at the expense of generation diversity, limiting the utility of these models as a source of synthetic training data. Although guidance-based approaches have been introduced to improve the utility of generated data by focusing on quality or diversity, the (implicit or explicit) utility functions oftentimes disregard the potential distribution shift between synthetic and real data. In this work, we introduce Chamfer Guidance: a training-free guidance approach which leverages a handful of real exemplar images to characterize the quality and diversity of synthetic data. We show that by leveraging the proposed Chamfer Guidance, we can boost the diversity of the generations w.r.t. a dataset of real images while maintaining or improving the generation quality on ImageNet-1k and standard geo-diversity benchmarks. Our approach achieves state-of-the-art few-shot performance with as little as 2 exemplar real images, obtaining 96.4% in terms of precision, and 86.4% in terms of distributional coverage, which increase to 97.5% and 92.7%, respectively, when using 32 real images. We showcase the benefits of the Chamfer Guidance generation by training downstream image classifiers on synthetic data, achieving accuracy boost of up to 15% for in-distribution over the baselines, and up to 16\% in out-of-distribution. Furthermore, our approach does not require using the unconditional model, and thus obtains a 31\% reduction in FLOPs w.r.t. classifier-free-guidance-based approaches at sampling time.

Lens flare occurs when shooting towards strong light sources, significantly degrading the visual quality of images. Due to the difficulty in capturing flare-corrupted and flare-free image pairs in the real world, existing datasets are typically synthesized in 2D by overlaying artificial flare templates onto background images. However, the lack of flare diversity in templates and the neglect of physical principles in the synthesis process hinder models trained on these datasets from generalizing well to real-world scenarios. To address these challenges, we propose a new physics-informed method for flare data generation, which consists of three stages: parameterized template creation, the laws of illumination-aware 2D synthesis, and physical engine-based 3D rendering, which finally gives us a mixed flare dataset that incorporates both 2D and 3D perspectives, namely FlareX. This dataset offers 9,500 2D templates derived from 95 flare patterns and 3,000 flare image pairs rendered from 60 3D scenes. Furthermore, we design a masking approach to obtain real-world flare-free images from their corrupted counterparts to measure the performance of the model on real-world images. Extensive experiments demonstrate the effectiveness of our method and dataset.