Rethinking Cross-Modal Interaction in Multimodal Diffusion Transformers

Multimodal Diffusion Transformers (MM-DiTs) have achieved remarkable progress in text-driven visual generation. However, even state-of-the-art MM-DiT models like FLUX struggle with achieving precise alignment between text prompts and generated content. We identify two key issues in the attention mechanism of MM-DiT, namely 1) the suppression of cross-modal attention due to token imbalance between visual and textual modalities and 2) the lack of timestep-aware attention weighting, which hinder the alignment. To address these issues, we propose \textbf{Temperature-Adjusted Cross-modal Attention (TACA)}, a parameter-efficient method that dynamically rebalances multimodal interactions through temperature scaling and timestep-dependent adjustment. When combined with LoRA fine-tuning, TACA significantly enhances text-image alignment on the T2I-CompBench benchmark with minimal computational overhead. We tested TACA on state-of-the-art models like FLUX and SD3.5, demonstrating its ability to improve image-text alignment in terms of object appearance, attribute binding, and spatial relationships. Our findings highlight the importance of balancing cross-modal attention in improving semantic fidelity in text-to-image diffusion models. Our codes are publicly available at \href{https://github.com/Vchitect/TACA}

Voyager: Long-Range and World-Consistent Video Diffusion for Explorable 3D Scene Generation

Real-world applications like video gaming and virtual reality often demand the ability to model 3D scenes that users can explore along custom camera trajectories. While significant progress has been made in generating 3D objects from text or images, creating long-range, 3D-consistent, explorable 3D scenes remains a complex and challenging problem. In this work, we present Voyager, a novel video diffusion framework that generates world-consistent 3D point-cloud sequences from a single image with user-defined camera path. Unlike existing approaches, Voyager achieves end-to-end scene generation and reconstruction with inherent consistency across frames, eliminating the need for 3D reconstruction pipelines (e.g., structure-from-motion or multi-view stereo). Our method integrates three key components: 1) World-Consistent Video Diffusion: A unified architecture that jointly generates aligned RGB and depth video sequences, conditioned on existing world observation to ensure global coherence 2) Long-Range World Exploration: An efficient world cache with point culling and an auto-regressive inference with smooth video sampling for iterative scene extension with context-aware consistency, and 3) Scalable Data Engine: A video reconstruction pipeline that automates camera pose estimation and metric depth prediction for arbitrary videos, enabling large-scale, diverse training data curation without manual 3D annotations. Collectively, these designs result in a clear improvement over existing methods in visual quality and geometric accuracy, with versatile applications.

MIRAGE: Assessing Hallucination in Multimodal Reasoning Chains of MLLM

Multimodal hallucination in multimodal large language models (MLLMs) restricts the correctness of MLLMs. However, multimodal hallucinations are multi-sourced and arise from diverse causes. Existing benchmarks fail to adequately distinguish between perception-induced hallucinations and reasoning-induced hallucinations. This failure constitutes a significant issue and hinders the diagnosis of multimodal reasoning failures within MLLMs. To address this, we propose the {\dataset} benchmark, which isolates reasoning hallucinations by constructing questions where input images are correctly perceived by MLLMs yet reasoning errors persist. {\dataset} introduces multi-granular evaluation metrics: accuracy, factuality, and LLMs hallucination score for hallucination quantification. Our analysis reveals that (1) the model scale, data scale, and training stages significantly affect the degree of logical, fabrication, and factual hallucinations; (2) current MLLMs show no effective improvement on spatial hallucinations caused by misinterpreted spatial relationships, indicating their limited visual reasoning capabilities; and (3) question types correlate with distinct hallucination patterns, highlighting targeted challenges and potential mitigation strategies. To address these challenges, we propose {\method}, a method that combines curriculum reinforcement fine-tuning to encourage models to generate logic-consistent reasoning chains by stepwise reducing learning difficulty, and collaborative hint inference to reduce reasoning complexity. {\method} establishes a baseline on {\dataset}, and reduces the logical hallucinations in original base models.

LabUtopia: High-Fidelity Simulation and Hierarchical Benchmark for Scientific Embodied Agents

Scientific embodied agents play a crucial role in modern laboratories by automating complex experimental workflows. Compared to typical household environments, laboratory settings impose significantly higher demands on perception of physical-chemical transformations and long-horizon planning, making them an ideal testbed for advancing embodied intelligence. However, its development has been long hampered by the lack of suitable simulator and benchmarks. In this paper, we address this gap by introducing LabUtopia, a comprehensive simulation and benchmarking suite designed to facilitate the development of generalizable, reasoning-capable embodied agents in laboratory settings. Specifically, it integrates i) LabSim, a high-fidelity simulator supporting multi-physics and chemically meaningful interactions; ii) LabScene, a scalable procedural generator for diverse scientific scenes; and iii) LabBench, a hierarchical benchmark spanning five levels of complexity from atomic actions to long-horizon mobile manipulation. LabUtopia supports 30 distinct tasks and includes more than 200 scene and instrument assets, enabling large-scale training and principled evaluation in high-complexity environments. We demonstrate that LabUtopia offers a powerful platform for advancing the integration of perception, planning, and control in scientific-purpose agents and provides a rigorous testbed for exploring the practical capabilities and generalization limits of embodied intelligence in future research.

Point-RFT: Improving Multimodal Reasoning with Visually Grounded Reinforcement Finetuning

Recent advances in large language models have significantly improved textual reasoning through the effective use of Chain-of-Thought (CoT) and reinforcement learning. However, extending these successes to vision-language tasks remains challenging due to inherent limitations in text-only CoT, such as visual hallucinations and insufficient multimodal integration. In this paper, we introduce Point-RFT, a multimodal reasoning framework explicitly designed to leverage visually grounded CoT reasoning for visual document understanding. Our approach consists of two stages: First, we conduct format finetuning using a curated dataset of 71K diverse visual reasoning problems, each annotated with detailed, step-by-step rationales explicitly grounded to corresponding visual elements. Second, we employ reinforcement finetuning targeting visual document understanding. On ChartQA, our approach improves accuracy from 70.88% (format-finetuned baseline) to 90.04%, surpassing the 83.92% accuracy achieved by reinforcement finetuning relying solely on text-based CoT. The result shows that our grounded CoT is more effective for multimodal reasoning compared with the text-only CoT. Moreover, Point-RFT exhibits superior generalization capability across several out-of-domain visual document reasoning benchmarks, including CharXiv, PlotQA, IconQA, TabMWP, etc., and highlights its potential in complex real-world scenarios.

Multi-Timescale Motion-Decoupled Spiking Transformer for Audio-Visual Zero-Shot Learning

Audio-visual zero-shot learning (ZSL) has been extensively researched for its capability to classify video data from unseen classes during training. Nevertheless, current methodologies often struggle with background scene biases and inadequate motion detail. This paper proposes a novel dual-stream Multi-Timescale Motion-Decoupled Spiking Transformer (MDST++), which decouples contextual semantic information and sparse dynamic motion information. The recurrent joint learning unit is proposed to extract contextual semantic information and capture joint knowledge across various modalities to understand the environment of actions. By converting RGB images to events, our method captures motion information more accurately and mitigates background scene biases. Moreover, we introduce a discrepancy analysis block to model audio motion information. To enhance the robustness of SNNs in extracting temporal and motion cues, we dynamically adjust the threshold of Leaky Integrate-and-Fire neurons based on global motion and contextual semantic information. Our experiments validate the effectiveness of MDST++, demonstrating their consistent superiority over state-of-the-art methods on mainstream benchmarks. Additionally, incorporating motion and multi-timescale information significantly improves HM and ZSL accuracy by 26.2\% and 39.9\%.

Decoupled Visual Interpretation and Linguistic Reasoning for Math Problem Solving

Current large vision-language models (LVLMs) typically employ a connector module to link visual features with text embeddings of large language models (LLMs) and use end-to-end training to achieve multi-modal understanding in a unified process. Well alignment needs high-quality pre-training data and a carefully designed training process. Current LVLMs face challenges when addressing complex vision-language reasoning tasks, with their reasoning capabilities notably lagging behind those of LLMs. This paper proposes a paradigm shift: instead of training end-to-end vision-language reasoning models, we advocate for developing a decoupled reasoning framework based on existing visual interpretation specialists and text-based reasoning LLMs. Our approach leverages (1) a dedicated vision-language model to transform the visual content of images into textual descriptions and (2) an LLM to perform reasoning according to the visual-derived text and the original question. This method presents a cost-efficient solution for multi-modal model development by optimizing existing models to work collaboratively, avoiding end-to-end development of vision-language models from scratch. By transforming images into language model-compatible text representations, it facilitates future low-cost and flexible upgrades to upcoming powerful LLMs. We introduce an outcome-rewarded joint-tuning strategy to optimize the cooperation between the visual interpretation and linguistic reasoning model. Evaluation results on vision-language benchmarks demonstrate that the decoupled reasoning framework outperforms recent LVLMs. Our approach yields particularly significant performance gains on visually intensive geometric mathematics problems. The code is available: https://github.com/guozix/DVLR.

High-Frequency Prior-Driven Adaptive Masking for Accelerating Image Super-Resolution

The primary challenge in accelerating image super-resolution lies in reducing computation while maintaining performance and adaptability. Motivated by the observation that high-frequency regions (e.g., edges and textures) are most critical for reconstruction, we propose a training-free adaptive masking module for acceleration that dynamically focuses computation on these challenging areas. Specifically, our method first extracts high-frequency components via Gaussian blur subtraction and adaptively generates binary masks using K-means clustering to identify regions requiring intensive processing. Our method can be easily integrated with both CNNs and Transformers. For CNN-based architectures, we replace standard $3 \times 3$ convolutions with an unfold operation followed by $1 \times 1$ convolutions, enabling pixel-wise sparse computation guided by the mask. For Transformer-based models, we partition the mask into non-overlapping windows and selectively process tokens based on their average values. During inference, unnecessary pixels or windows are pruned, significantly reducing computation. Moreover, our method supports dilation-based mask adjustment to control the processing scope without retraining, and is robust to unseen degradations (e.g., noise, compression). Extensive experiments on benchmarks demonstrate that our method reduces FLOPs by 24--43% for state-of-the-art models (e.g., CARN, SwinIR) while achieving comparable or better quantitative metrics. The source code is available at https://github.com/shangwei5/AMSR

Bridging Geometry-Coherent Text-to-3D Generation with Multi-View Diffusion Priors and Gaussian Splatting

Score Distillation Sampling (SDS) leverages pretrained 2D diffusion models to advance text-to-3D generation but neglects multi-view correlations, being prone to geometric inconsistencies and multi-face artifacts in the generated 3D content. In this work, we propose Coupled Score Distillation (CSD), a framework that couples multi-view joint distribution priors to ensure geometrically consistent 3D generation while enabling the stable and direct optimization of 3D Gaussian Splatting. Specifically, by reformulating the optimization as a multi-view joint optimization problem, we derive an effective optimization rule that effectively couples multi-view priors to guide optimization across different viewpoints while preserving the diversity of generated 3D assets. Additionally, we propose a framework that directly optimizes 3D Gaussian Splatting (3D-GS) with random initialization to generate geometrically consistent 3D content. We further employ a deformable tetrahedral grid, initialized from 3D-GS and refined through CSD, to produce high-quality, refined meshes. Quantitative and qualitative experimental results demonstrate the efficiency and competitive quality of our approach.

EarthMapper: Visual Autoregressive Models for Controllable Bidirectional Satellite-Map Translation

Satellite imagery and maps, as two fundamental data modalities in remote sensing, offer direct observations of the Earth's surface and human-interpretable geographic abstractions, respectively. The task of bidirectional translation between satellite images and maps (BSMT) holds significant potential for applications in urban planning and disaster response. However, this task presents two major challenges: first, the absence of precise pixel-wise alignment between the two modalities substantially complicates the translation process; second, it requires achieving both high-level abstraction of geographic features and high-quality visual synthesis, which further elevates the technical complexity. To address these limitations, we introduce EarthMapper, a novel autoregressive framework for controllable bidirectional satellite-map translation. EarthMapper employs geographic coordinate embeddings to anchor generation, ensuring region-specific adaptability, and leverages multi-scale feature alignment within a geo-conditioned joint scale autoregression (GJSA) process to unify bidirectional translation in a single training cycle. A semantic infusion (SI) mechanism is introduced to enhance feature-level consistency, while a key point adaptive guidance (KPAG) mechanism is proposed to dynamically balance diversity and precision during inference. We further contribute CNSatMap, a large-scale dataset comprising 302,132 precisely aligned satellite-map pairs across 38 Chinese cities, enabling robust benchmarking. Extensive experiments on CNSatMap and the New York dataset demonstrate EarthMapper's superior performance, achieving significant improvements in visual realism, semantic consistency, and structural fidelity over state-of-the-art methods. Additionally, EarthMapper excels in zero-shot tasks like in-painting, out-painting and coordinate-conditional generation, underscoring its versatility.