Training-Free Sparse Attention for Fast Video Generation via Offline Layer-Wise Sparsity Profiling and Online Bidirectional Co-Clustering

Diffusion Transformers (DiTs) achieve strong video generation quality but suffer from high inference cost due to dense 3D attention, leading to the development of sparse attention technologies to improve efficiency. However, existing training-free sparse attention methods in video generation still face two unresolved limitations: ignoring layer heterogeneity in attention pruning and ignoring query-key coupling in block partitioning, which hinder a better quality-speedup trade-off. In this work, we uncover a critical insight that the attention sparsity of each layer is its intrinsic property, with minor effects across different inputs. Motivated by this, we propose SVOO, a training-free Sparse attention framework for fast Video generation via Offline layer-wise sparsity profiling and Online bidirectional co-clustering. Specifically, SVOO adopts a two-stage paradigm: (i) offline layer-wise sensitivity profiling to derive intrinsic per-layer pruning levels, and (ii) online block-wise sparse attention via a novel bidirectional co-clustering algorithm. Extensive experiments on seven widely used video generation models demonstrate that SVOO achieves a superior quality-speedup trade-off over state-of-the-art methods, delivering up to $1.93\times$ speedup while maintaining a PSNR of up to 29 dB on Wan2.1.

PhysVideo: Physically Plausible Video Generation with Cross-View Geometry Guidance

Recent progress in video generation has led to substantial improvements in visual fidelity, yet ensuring physically consistent motion remains a fundamental challenge. Intuitively, this limitation can be attributed to the fact that real-world object motion unfolds in three-dimensional space, while video observations provide only partial, view-dependent projections of such dynamics. To address these issues, we propose PhysVideo, a two-stage framework that first generates physics-aware orthogonal foreground videos and then synthesizes full videos with background. In the first stage, Phys4View leverages physics-aware attention to capture the influence of physical attributes on motion dynamics, and enhances spatio-temporal consistency by incorporating geometry-enhanced cross-view attention and temporal attention. In the second stage, VideoSyn uses the generated foreground videos as guidance and learns the interactions between foreground dynamics and background context for controllable video synthesis. To support training, we construct PhysMV, a dataset containing 40K scenes, each consisting of four orthogonal viewpoints, resulting in a total of 160K video sequences. Extensive experiments demonstrate that PhysVideo significantly improves physical realism and spatial-temporal coherence over existing video generation methods. Home page: https://anonymous.4open.science/w/Phys4D/.

VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model

Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather than action-relevant state transitions, making them vulnerable to appearance bias, nuisance motion, and information leakage. We introduce VLA-JEPA, a JEPA-style pretraining framework that sidesteps these pitfalls by design. The key idea is leakage-free state prediction: a target encoder produces latent representations from future frames, while the student pathway sees only the current observation -- future information is used solely as supervision targets, never as input. By predicting in latent space rather than pixel space, VLA-JEPA learns dynamics abstractions that are robust to camera motion and irrelevant background changes. This yields a simple two-stage recipe -- JEPA pretraining followed by action-head fine-tuning -- without the multi-stage complexity of prior latent-action pipelines. Experiments on LIBERO, LIBERO-Plus, SimplerEnv and real-world manipulation tasks show that VLA-JEPA achieves consistent gains in generalization and robustness over existing methods.

WorldArena: A Unified Benchmark for Evaluating Perception and Functional Utility of Embodied World Models

While world models have emerged as a cornerstone of embodied intelligence by enabling agents to reason about environmental dynamics through action-conditioned prediction, their evaluation remains fragmented. Current evaluation of embodied world models has largely focused on perceptual fidelity (e.g., video generation quality), overlooking the functional utility of these models in downstream decision-making tasks. In this work, we introduce WorldArena, a unified benchmark designed to systematically evaluate embodied world models across both perceptual and functional dimensions. WorldArena assesses models through three dimensions: video perception quality, measured with 16 metrics across six sub-dimensions; embodied task functionality, which evaluates world models as data engines, policy evaluators, and action planners integrating with subjective human evaluation. Furthermore, we propose EWMScore, a holistic metric integrating multi-dimensional performance into a single interpretable index. Through extensive experiments on 14 representative models, we reveal a significant perception-functionality gap, showing that high visual quality does not necessarily translate into strong embodied task capability. WorldArena benchmark with the public leaderboard is released at https://worldarena.ai, providing a framework for tracking progress toward truly functional world models in embodied AI.

Back to Physics: Operator-Guided Generative Paths for SMS MRI Reconstruction

Simultaneous multi-slice (SMS) imaging with in-plane undersampling enables highly accelerated MRI but yields a strongly coupled inverse problem with deterministic inter-slice interference and missing k-space data. Most diffusion-based reconstructions are formulated around Gaussian-noise corruption and rely on additional consistency steps to incorporate SMS physics, which can be mismatched to the operator-governed degradations in SMS acquisition. We propose an operator-guided framework that models the degradation trajectory using known acquisition operators and inverts this process via deterministic updates. Within this framework, we introduce an operator-conditional dual-stream interaction network (OCDI-Net) that explicitly disentangles target-slice content from inter-slice interference and predicts structured degradations for operator-aligned inversion, and we instantiate reconstruction as a two-stage chained inference procedure that performs SMS slice separation followed by in-plane completion. Experiments on fastMRI brain data and prospectively acquired in vivo diffusion MRI data demonstrate improved fidelity and reduced slice leakage over conventional and learning-based SMS reconstructions.

TrackVLA++: Unleashing Reasoning and Memory Capabilities in VLA Models for Embodied Visual Tracking

Embodied Visual Tracking (EVT) is a fundamental ability that underpins practical applications, such as companion robots, guidance robots and service assistants, where continuously following moving targets is essential. Recent advances have enabled language-guided tracking in complex and unstructured scenes. However, existing approaches lack explicit spatial reasoning and effective temporal memory, causing failures under severe occlusions or in the presence of similar-looking distractors. To address these challenges, we present TrackVLA++, a novel Vision-Language-Action (VLA) model that enhances embodied visual tracking with two key modules, a spatial reasoning mechanism and a Target Identification Memory (TIM). The reasoning module introduces a Chain-of-Thought paradigm, termed Polar-CoT, which infers the target's relative position and encodes it as a compact polar-coordinate token for action prediction. Guided by these spatial priors, the TIM employs a gated update strategy to preserve long-horizon target memory, ensuring spatiotemporal consistency and mitigating target loss during extended occlusions. Extensive experiments show that TrackVLA++ achieves state-of-the-art performance on public benchmarks across both egocentric and multi-camera settings. On the challenging EVT-Bench DT split, TrackVLA++ surpasses the previous leading approach by 5.1 and 12, respectively. Furthermore, TrackVLA++ exhibits strong zero-shot generalization, enabling robust real-world tracking in dynamic and occluded scenarios.

Comp-X: On Defining an Interactive Learned Image Compression Paradigm With Expert-driven LLM Agent

We present Comp-X, the first intelligently interactive image compression paradigm empowered by the impressive reasoning capability of large language model (LLM) agent. Notably, commonly used image codecs usually suffer from limited coding modes and rely on manual mode selection by engineers, making them unfriendly for unprofessional users. To overcome this, we advance the evolution of image coding paradigm by introducing three key innovations: (i) multi-functional coding framework, which unifies different coding modes of various objective/requirements, including human-machine perception, variable coding, and spatial bit allocation, into one framework. (ii) interactive coding agent, where we propose an augmented in-context learning method with coding expert feedback to teach the LLM agent how to understand the coding request, mode selection, and the use of the coding tools. (iii) IIC-bench, the first dedicated benchmark comprising diverse user requests and the corresponding annotations from coding experts, which is systematically designed for intelligently interactive image compression evaluation. Extensive experimental results demonstrate that our proposed Comp-X can understand the coding requests efficiently and achieve impressive textual interaction capability. Meanwhile, it can maintain comparable compression performance even with a single coding framework, providing a promising avenue for artificial general intelligence (AGI) in image compression.

Structure-preserving Feature Alignment for Old Photo Colorization

Deep learning techniques have made significant advancements in reference-based colorization by training on large-scale datasets. However, directly applying these methods to the task of colorizing old photos is challenging due to the lack of ground truth and the notorious domain gap between natural gray images and old photos. To address this issue, we propose a novel CNN-based algorithm called SFAC, i.e., Structure-preserving Feature Alignment Colorizer. SFAC is trained on only two images for old photo colorization, eliminating the reliance on big data and allowing direct processing of the old photo itself to overcome the domain gap problem. Our primary objective is to establish semantic correspondence between the two images, ensuring that semantically related objects have similar colors. We achieve this through a feature distribution alignment loss that remains robust to different metric choices. However, utilizing robust semantic correspondence to transfer color from the reference to the old photo can result in inevitable structure distortions. To mitigate this, we introduce a structure-preserving mechanism that incorporates a perceptual constraint at the feature level and a frozen-updated pyramid at the pixel level. Extensive experiments demonstrate the effectiveness of our method for old photo colorization, as confirmed by qualitative and quantitative metrics.

LiftVSR: Lifting Image Diffusion to Video Super-Resolution via Hybrid Temporal Modeling with Only 4$\times$RTX 4090s

Diffusion models have significantly advanced video super-resolution (VSR) by enhancing perceptual quality, largely through elaborately designed temporal modeling to ensure inter-frame consistency. However, existing methods usually suffer from limited temporal coherence and prohibitively high computational costs (e.g., typically requiring over 8 NVIDIA A100-80G GPUs), especially for long videos. In this work, we propose LiftVSR, an efficient VSR framework that leverages and elevates the image-wise diffusion prior from PixArt-$\alpha$, achieving state-of-the-art results using only 4$\times$RTX 4090 GPUs. To balance long-term consistency and efficiency, we introduce a hybrid temporal modeling mechanism that decomposes temporal learning into two complementary components: (i) Dynamic Temporal Attention (DTA) for fine-grained temporal modeling within short frame segment ($\textit{i.e.}$, low complexity), and (ii) Attention Memory Cache (AMC) for long-term temporal modeling across segments ($\textit{i.e.}$, consistency). Specifically, DTA identifies multiple token flows across frames within multi-head query and key tokens to warp inter-frame contexts in the value tokens. AMC adaptively aggregates historical segment information via a cache unit, ensuring long-term coherence with minimal overhead. To further stabilize the cache interaction during inference, we introduce an asymmetric sampling strategy that mitigates feature mismatches arising from different diffusion sampling steps. Extensive experiments on several typical VSR benchmarks have demonstrated that LiftVSR achieves impressive performance with significantly lower computational costs.

Why Compress What You Can Generate? When GPT-4o Generation Ushers in Image Compression Fields

The rapid development of AIGC foundation models has revolutionized the paradigm of image compression, which paves the way for the abandonment of most pixel-level transform and coding, compelling us to ask: why compress what you can generate if the AIGC foundation model is powerful enough to faithfully generate intricate structure and fine-grained details from nothing more than some compact descriptors, i.e., texts, or cues. Fortunately, recent GPT-4o image generation of OpenAI has achieved impressive cross-modality generation, editing, and design capabilities, which motivates us to answer the above question by exploring its potential in image compression fields. In this work, we investigate two typical compression paradigms: textual coding and multimodal coding (i.e., text + extremely low-resolution image), where all/most pixel-level information is generated instead of compressing via the advanced GPT-4o image generation function. The essential challenge lies in how to maintain semantic and structure consistency during the decoding process. To overcome this, we propose a structure raster-scan prompt engineering mechanism to transform the image into textual space, which is compressed as the condition of GPT-4o image generation. Extensive experiments have shown that the combination of our designed structural raster-scan prompts and GPT-4o's image generation function achieved the impressive performance compared with recent multimodal/generative image compression at ultra-low bitrate, further indicating the potential of AIGC generation in image compression fields.