Denoising
-
Reflex: Real-Time VLA Control through Streaming Inference
2026-07-16 07:56:43Yuanchun Guo, Bingyan LiuAbstract
Flow matching Vision-Language-Action (VLA) models promise precise continuous control, but their iterative denoising nature introduces fundamental incompatibilities with real-time robotics: global timestep injection invalidates KV-caching, forcing a choice between slow $O(N^2)$ re-computation or mathematically incorrect cache reuse. We present \textbf{Reflex}, a framework that enables \textit{real-time streaming inference} for flow matching policies by exploiting the \textit{Timestep-Invariance Property} -- that perception encoders are functionally independent of the denoising loop. Reflex partitions the attention context into static, sliding, and dynamic regions, enabling $O(1)$ incremental cache updates while preserving full-batch-equivalent attention outputs for fixed inputs. To ensure stability under continuous high-frequency inference, we introduce \textit{AdaRMSNorm}, an adaptive normalization layer that prevents BFloat16 numerical collapse by gating on flow phase. We further maximize throughput through an \textit{async pipeline} that decouples visual encoding from action generation, combined with \textit{operator fusion} that reduces kernel overhead. On LIBERO and Kinetix benchmarks, Reflex achieves a 2.58$\times$ inference speedup and 50Hz stable streaming, reducing reaction latency by up to 54\% and enabling efficient deployment without performance degradation.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14695
PDF
-
NavCMPO: Critic-Guided MeanFlow Policy Optimization for Adaptive Navigation
2026-07-16 07:10:46Junjie An, Yi Wu, Xiao Liu, Yiqun Zhou, Yuechen Wu, Xiaoqing Guan, You Wang, Guang LiAbstract
End-to-end diffusion-based policies have demonstrated strong performance in mapless visual navigation, but their iterative denoising process introduces substantial inference latency, while behavior cloning limits performance to the quality of expert demonstrations. We present NavCMPO, a two-stage adaptive navigation framework that combines few-step MeanFlow trajectory generation, critic-guided refinement, and reinforcement learning fine-tuning. During pre-training, an obstacle proximity prediction task encourages the visual representation to capture obstacle-aware spatial information. To compensate for the degradation in obstacle avoidance caused by few-step generation, Critic-Guided Trajectory Refinement (CGTR) uses gradients from a critic trained with obstacle-point-cloud supervision to refine intermediate trajectories. During adaptation, the MeanFlow policy is fine-tuned using Proximal Policy Optimization with behavior-cloning regularization, while the critic is updated to accommodate embodiment-specific observation changes. Under a matched training budget on the InternVLA-N1 benchmark, NavCMPO achieves an average success rate of 74.7\%, exceeding the retrained NavDP baseline by 6.4 percentage points, while reducing inference latency from 85\,ms to 60\,ms. Experiments on a Unitree Go2 further demonstrate effective sim-to-real transfer.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14643
PDF
-
Seeing the End at Step Zero: Accelerating Diffusion MLLMs via MLP Sparsity-Aware Truncation
2026-07-16 04:37:02Qicheng Zhao, Qi Sun, Zheyu YanAbstract
Diffusion Multimodal Large Language Models (DMLLMs) are highly effective for multimodal reasoning, yet their inference efficiency is significantly hindered by fixed-length generation constraints. Since the actual output length is unknown, output sequences are padded to a predefined maximum length, resulting in substantial redundant computation over unnecessary [EOS] tokens. In this work, we discover that DMLLMs implicitly reveal their valid semantic boundary at the very first denoising step through a distinct shift in MLP activation sparsity. Leveraging this observation, we propose Seer, a training-free framework that detects this boundary using a Signal-to-Noise Ratio (SNR)-based criterion and performs one-shot truncation of the redundant suffix for all subsequent computations. To preserve these theoretical gains during batched serving, Seer incorporates a hybrid execution strategy that maximizes throughput while seamlessly accommodating dynamic sequence lengths. Experimental results demonstrate that Seer effectively eliminates padding waste, accelerating throughput by up to $\sim$31$\times$. Across 9 benchmarks, Seer robustly maintains overall performance and even improves accuracy on complex visual tasks by mitigating noise leakage (e.g., DocVQA score increases from 63.52 to 63.66), offering a highly efficient, plug-and-play solution for DMLLM acceleration.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14557
PDF
-
Motion Planning with Model-Based Diffusion via Constraint Optimization and Adaptive Scheduling
2026-07-16 01:04:32Zhilin He, Bowei Li, Jianlin Dou, Yuner Zhang, Changliu LiuAbstract
Single-Robot Motion Planning (SRMP) in highly non-convex constrained environments, where robots must satisfy collision-free guarantees, dynamic feasibility, and task-related constraints, is challenging under complex constraints and computational limits. Recent Model-Based Diffusion (MBD) approaches recast the SRMP as trajectory optimization that samples from a posterior over trajectories, using known dynamics, and analytically estimates the score function from rollout samples to guide diffusion denoising toward a low-cost, clean trajectory without demonstration learning. While existing works further adapt MBD to constrained environments and showcase promising performance, they are still limited by (1) enforcing safety either via soft feasibility diffusion priors or hard projection operators, but lack a unified framework to integrate both, and (2) fixing safety enforcement to neglect the changing of diffusion scheduling. Therefore, we introduce Model-Based Diffusion via Constraint Optimization and Adaptive Scheduling (MD-COAS) for SRMP that unifies the inexact Augmented Lagrangian Method (iALM) soft diffusion prior with a Convex Feasible Set (CFS)-based hard projection operator, and adaptively schedules and co-optimizes safety enforcement, along with diffusion scheduling. Experiments demonstrate that our method achieves higher safety \& success rates, faster convergence, and lower final costs than baseline planners on randomly generated highly non-convex 2D benchmarks and a 7-DoF robot arm avoidance task.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14455
PDF
-
Integration Matters: Rollout-Based Training for Constrained Diffusion Models
2026-07-15 22:25:59Xiaoxuan Liang, Saeid Naderiparizi, Berend Zwartsenberg, Frank WoodAbstract
Constrained generative models aim to produce samples that satisfy complex feasibility constraints while remaining faithful to the data distribution. Existing constrained generation methods typically enforce constraints either through training-time optimization or sampling-time correction. Training-time optimization approaches optimize on states induced by the training distribution, which can differ substantially from those encountered during sampling. Sampling-time correction methods instead modify the sampling process at inference, introducing distribution shift and requiring expensive tuning, particularly for few-step sampling. We propose a fine-tuning framework that incorporates constraint guidance obtained through online rollout into the training process, which aligns training with sampling by differentiating through the fixed noise schedule used to numerically integrate the denoising process. This exposes the model to violations that arise along the denoising trajectory and aligns diffusion learning with the sampling process. Experiments across multiple tasks show that our method improves constraint satisfaction while maintaining competitive sampling quality compared to prior methods.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14398
PDF
-
M$^text{4}$World: A Multi-view Multimodal Driving World Model for Interactive Object Manipulation and Minute-long Streaming
2026-07-15 16:36:33Ke Cheng, Hanqiao Ye, Lei Shi, Yahui Liu, Yunhan Shen, Jingtao Dong, Zhenke Wang, Wenxuan Ao, Weixiang Xu, Kaining Huang, Shuhan ShenAbstract
Driving-world generation has emerged as a core capability for scalable autonomous-driving simulation, yet existing methods remain limited in object-level controllability and long-horizon stability. We present M$^\text{4}$World, a Multi-view and Multimodal generative driving world model that synthesizes future surround-view video streams and synchronized LiDAR scans while supporting interactive object Manipulation and stable Minute-long streaming. Fine-grained object manipulation is realized through a flexible conditioning interface that supports explicit control over both the spatial layout and visual appearance of individual objects. Stable minute-long streaming, on the other hand, is achieved through a multi-stage training framework that enables online causal generation in only four denoising steps while maintaining coherent world dynamics throughout extended rollouts. Building on these components, we introduce an efficient few-clip post-training as well as a suite of visual reference-conditioned generation models, preserving general generation ability while allowing rare-case customization for long-tail controllability. To assess controllability beyond realism, we further introduce an automated VLM-based judging pipeline that evaluates scene-level condition adherence, view-wise object controllability, and cross-view object consistency. Comprehensive experiments show that M$^\text{4}$World consistently delivers high generation quality, precise controllability, and stable minute-long streaming. Together with downstream long-tail augmentation and scene editing, these results demonstrate the potential of M$^\text{4}$World for controllable, scalable driving simulation.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14005
PDF
-
Inference-Time Concept Suppression and Video-Centric Evaluation for Text-to-Video Models
Abstract
Text-to-video (T2V) generators can synthesize realistic and temporally coherent videos, but controllably removing a target concept from a generator remains difficult. Unlike text-to-image concept erasure, T2V unlearning must suppress a target concept that may persist across frames while preserving non-target subjects, actions, scenes, and temporal structure. We propose \textbf{SIRUS}, a training-free inference-time framework for concept-level T2V unlearning. Given textual aliases of a target concept, SIRUS localizes target-related prompt evidence and suppresses target expression during sampling, without updating the text encoder or denoising network. We further introduce a video-oriented evaluation framework for T2V unlearning that separately measures target forgetting, non-target preservation, video quality, jailbreak robustness, and efficiency, using video-level failure criteria, frame-level residue statistics, paired preservation analysis, VBench-based quality diagnostics, and deployment overhead measurement. Across five safety, object, and style concepts on CogVideoX, SIRUS reaches 70.4\% average forgetting success and 25.7\% average frame hit, compared with 44.4\% / 47.2\% for VideoEraser, while reducing the average VBench quality drop from -0.043 to -0.016, yielding the strongest forgetting-quality trade-off among fully evaluated baselines. Transfer experiments on Wan2.2 further suggest that SIRUS generalizes across modern T2V backbones.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14194
PDF
-
DNA: Dual-stage Native Attribution for Generated Image Source Tracing
2026-07-15 10:27:52Chao Wang, Kejiang Chen, Zijin Yang, Yaofei Wang, Yuang Qi, Weiming Zhang, Nenghai YuAbstract
The rapid evolution of image generation has produced numerous within-family variants, making source-model attribution of suspect images increasingly important for digital forensics. Existing proactive methods rely on watermark embedding or model modification, which may degrade visual quality and limit deployment flexibility. Passive methods often rely on large-scale supervised training or a single reconstruction signal, limiting their ability to handle unknown sources and distinguish highly similar within-family variants. We observe that attribution signals in latent generative models are naturally stratified across architectural levels: VAE-level cues reflect family-shared information, whereas backbone-level cues capture variant-specific behaviors. Motivated by this insight, we propose Dual-stage Native Attribution (DNA), a coarse-to-fine framework that follows this hierarchy without additional neural-network training. The coarse-grained stage uses Autoencoder Double-Reconstruction (AEDR) for efficient open-set family-level screening. The fine-grained stage performs closed-set model-level attribution with Native Prediction Consistency (NPC), which compares native prediction errors of within-family variants across multiple noise levels under semantic conditioning and attributes the source via normalized calibrated scores. To enable systematic evaluation, we construct DNA-30K, a benchmark for within-family variant attribution under open-set family-level evaluation. It comprises 30,000 images generated by 24 candidate models across six families spanning both denoising diffusion and flow matching, plus non-candidate generated and natural images as unknown sources. Experiments show that DNA achieves 89.11% end-to-end attribution accuracy on a task where random guessing accuracy is below 1% and outperforms the strongest baseline by 33.81% even when AEDR is used as the coarse-grained stage.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13685
PDF
-
The Devil is in the Dark Pixels: Toward Brightness Bias-Robust Denoising
2026-07-15 08:57:38Sungjun Cho, Zhuangzhuang Chen, Xiaomeng LiAbstract
In this paper, we reveal an important yet overlooked problem in image denoising: under signal-dependent camera noise models, dark regions suffer from inherently low Signal-to-Noise Ratio (SNR), as signal intensity decays far faster than noise variance diminishes, making detail recovery in dark areas fundamentally challenging. Yet rather than compensating for this difficulty, MSE-trained denoisers exacerbate it -- reconstructing dark pixels up to 6x worse relative to their per-band noise floor. This bias stems from two compounding factors: signal-dependent noise inflates bright-pixel residuals, and the network's Jacobian norm increases monotonically with brightness. Together, these cause bright regions to chronically dominate gradient updates at the expense of dark ones. To this end, we propose Brightness Bias-Robust Denoising (BBRD), a drop-in replacement for MSE loss that partitions pixels into brightness bands, normalizes per-band error by empirical noise variance, and applies Group Distributionally Robust Optimization (Group-DRO) to dynamically upweight whichever band is currently worst, with zero additional parameters or inference cost. Across 8 architectures and 2 datasets in our experiments, BBRD is the only method among 13 tested alternatives that improves each brightness band simultaneously, achieving up to +0.45 dB on dark bands, +0.32 dB on bright bands, and +0.65 dB aggregate Peak Signal-to-Noise Ratio (PSNR) on SIDD, with the largest per-band gains in the darkest regions where detail recovery matters most. Code is available at this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2607.16320
PDF
-
Optimizing Denoising Trajectories in dLLMs: A Lightweight Evolutionary Heuristic Approach
2026-07-15 07:52:50Zijian Zhao, Dian Jin, Xialiang Tong, Sen Li, Mingxuan YuanAbstract
Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to conventional Auto-Regressive (AR) Large Language Models (LLMs). By leveraging bidirectional attention and parallel decoding, dLLMs enable more efficient generation. However, they require a carefully designed denoising scheduler at inference time (absent during training) whose choice significantly impacts generation quality. While confidence-based heuristic schedulers have shown strong empirical performance, they suffer from two critical failure modes: EOS Overflow and Proximal Bias. Through in-depth analysis of the Transformer's attention patterns, we reveal that these failures stem from certain positions assigning disproportionately high attention weights to invalid tokens (e.g., [MASK] and [EOS]), which produce misleading confidence signals. Building on this insight, empirical evidence shows that valid attention scores can provide complementary guidance to conventional confidence-based heuristics, yet no single metric consistently excels across all scenarios, implying that the optimal denoising trajectory is highly context-dependent. To address this problem, we propose a lightweight evolutionary heuristic scheduler optimized using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). Our scheduler dynamically integrates multiple heuristic features with a contextual mean-field embedding, while requiring only 393 trainable parameters. Evaluated on LLaDA and Dream across four reasoning and planning benchmarks, our method consistently outperforms strong baselines, including conventional heuristics, block auto-regressive methods, and recent State-Of-The-Art (SOTA) approaches. To the best of our knowledge, it represents the most parameter-efficient neural scheduler to date. Our code is available at this https URL .
Abstract (translated)
URL
https://arxiv.org/abs/2609.26052
PDF
-
Discrete Diffusion Models: A Unified Framework from Tokenization to Generation
2026-07-15 04:23:22Ye Yuan, Weien Li, Rui Song, Zeyu Li, Haochen Liu, Xiangyu Kong, Zixuan Dong, Linfeng Du, Zipeng Sun, Weixu Zhang, Jiaxin Huang, Changjiang Han, Yonghan Yang, Zichen Zhao, Xiuyuan Hu, Haolun Wu, Yankai Chen, Fengran Mo, Jikun Kang, Bowei He, Philip S. Yu, Xue LiuAbstract
Discrete denoising diffusion models (DDMs) have recently emerged as a compelling alternative to autoregressive (AR) modeling for discrete data, offering parallel generation and iterative global refinement capabilities. Unlike continuous diffusion, where the state space is fixed, DDMs are fundamentally shaped by how the discrete state space is constructed: the tokenization scheme, the vocabulary topology, and domain-specific structural alphabets. This work introduces a unified conceptual framework that views discrete diffusion models through the construction of the underlying discrete state space. Within this framework, existing formulations, including transition-matrix, masking/absorbing-state, and score/ratio-based approaches, emerge as different instantiations of a common design space. The framework further exposes common design trade-offs across training objectives, inference algorithms, scaling behavior, systems optimization, and evaluation protocols, suggesting several promising directions for future research.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13431
PDF
-
WNOJ-LIO: A White-Noise-on-Jerk Motion-Prior EKF for High-Dynamic LiDAR-IMU Fusion
2026-07-15 03:07:44Junning Lyu, Qizhi Guo, Xia Ning, Tao Song, Shaoming HeAbstract
LiDAR-inertial odometry (LIO) is a key component of autonomous navigation, but high-dynamic driving exposes two coupled challenges: intra-scan motion distortion and vibration-contaminated inertial measurements. Most real-time LiDAR-inertial pipelines propagate the system state by integrating raw IMU measurements and then use the propagated trajectory for point cloud de-distortion, thereby propagating inertial noise into both the corrected scan and the subsequent scan-to-map registration. This paper presents WNOJ-LIO, a LiDAR-IMU fusion framework based on a White-Noise-on-Jerk (WNOJ) Extended Kalman Filter (EKF). WNOJ-LIO employs a decoupled WNOJ prior on $\R^3 \times \SO(3)$ for state prediction and treats the IMU as a high-frequency measurement source rather than the driver of state propagation. The resulting posterior state history is then used for LiDAR scan de-distortion and subsequent point-to-plane LiDAR updates. The decoupled process model enables closed-form covariance propagation, thereby bridging the gap between batch WNOJ Gaussian process (GP) trajectory priors and recursive filtering. Simulation results demonstrate improvements in acceleration and angular-velocity denoising, scan de-distortion, and localization accuracy over a FAST-LIO-style baseline. Real-world experiments were conducted using an autonomous racing car on four driving segments with maximum speeds ranging from 53 to 208~km/h, covering a wide range of vehicle vibration levels. The experiments further validate the proposed method and provide a comprehensive evaluation of its performance in estimating acceleration, angular velocity, body-frame linear velocity, attitude, and position under highly dynamic driving. The source code of WNOJ-LIO is publicly available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13405
PDF
-
Dose-Aware Cold Diffusion with Physics Consistency for Generalizable Low-Dose CT Reconstruction
2026-07-15 02:34:58Md Imam Ahasan, Guangchao Yang, A F M Abdun Noor, S M Hasan Mahmud, Md Mahfuzur RahmanAbstract
Reducing radiation dose in computed tomography significantly degrades image quality and poses challenges for accurate and clinically reliable reconstruction. While recent approaches have shown promise for low-dose CT, they often struggle to generalize across continuous and previously unseen dose levels, leading to artifacts and loss of anatomical detail. To address these limitations, we propose Dose-Aware Cold Diffusion (DACD), a physics-consistent reconstruction framework that explicitly models radiation dose as a continuous latent factor within a cold diffusion process. The proposed DACD framework integrates image-based dose-aware perception, multi-scale structural prior extraction, and dose-calibrated step allocation to adaptively guide the denoising trajectory. In addition, an iterative forward-backprojection correction is incorporated into the reverse refinement process to enforce projection-domain data consistency. Extensive experiments on three public benchmarks, including Mayo-2020, Mayo-2016, and LoDoPaB-CT, demonstrate that DACD consistently outperforms state-of-the-art diffusion-based and physics-guided methods in both quantitative accuracy and visual fidelity, particularly under ultra-low-dose conditions. The results show that DACD achieves robust generalization across a continuous range of dose levels, including those unseen during training.
Abstract (translated)
URL
https://arxiv.org/abs/2609.18943
PDF
-
Efficient Text-to-Audio Generation via Pruning
2026-07-14 23:22:27Arshdeep Singh, Yi Yuan, Yun Chen, Wenwu Wang, Mark D. PlumbleyAbstract
Diffusion-based text-to-audio generative models such as AudioLDM achieve high perceptual quality and strong semantic consistency; however, their practical deployment is hindered by the substantial computational cost of the U-Net denoising backbone. In this work, we apply model pruning to improve the computational efficiency of AudioLDM, a U-Net-based text-conditioned audio latent diffusion model. We analyse parameter redundancy across U-Net convolutional blocks and evaluate a filter-pruning strategy. Pruning is guided by norm-based criteria and followed by lightweight finetuning to recover performance losses. Experimental results demonstrate that up to 83% of the parameters and 39% of the multiply-accumulate operations of U-Net have been reduced while maintaining, and in some cases improving, generation quality compared to the baseline unpruned network. We find that pruning affects AudioLDM's ability to generate certain sound events including safety-critical sounds such as gunshots, sirens, and explosions, as well as mechanical sounds such as drills and sewing machines, and other sounds such as sprays and tick-tocks, which are mostly recovered by lightweight finetuning of the pruned model.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13330
PDF
-
Improving Medical Image Generative Models with Fr'echet Distance Loss
2026-07-14 22:11:15Andrew Marshall, Xuanang Xu, Xiaoran Zhang, Rui Wang, Lawrence Staib, James DuncanAbstract
Diffusion generative models have demonstrated immense potential for synthetic medical image generation. However, these models often struggle to capture complex morphological characteristics of heterogeneous tumors with irregular boundaries, limiting their utility for downstream clinical tasks such as segmentation. This limitation stems from the standard denoising objective: minimizing a per-pixel error, which smooths high-variance irregular structures characteristic of tumors. To address this, we propose finetuning these generative models with Fréchet Distance loss (FD-loss). FD-loss aligns the first and second order feature statistics of real and generated images in a pretrained encoder space, encouraging the generator to capture complex structural variations characteristic of heterogeneous tumors. We integrate FD-loss across diverse architectural settings, using both natural- and medical-image encoders on multiple liver and brain cancer datasets spanning CT and MRI modalities. Downstream segmentation networks trained on our FD-regularized synthetic data consistently achieve superior performance, improving tumor DSC by $>$$5\%$ over unregularized synthetic augmentation alone. Qualitative analysis suggests these gains are associated with more faithful tumor synthesis and fewer segmentation hallucinations. Our results show FD-loss as an effective regularizer for medical image generative models to improve clinical workflows.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13300
PDF
-
Adaptive Filtering of the KV Cache: Diagnosing and Correcting Structural-Role Bias in LLM Inference
Abstract
Attention-based KV cache eviction (H2O and its descendants) compresses the memory-constrained state of a long-context model by ranking tokens on accumulated attention mass, treated here as signal energy, and keeping the heaviest. On schema-dense input streams such as nested JSON, this score acts as a non-stationary filter that disproportionately retains noise: a non-content sink role (delimiters or whitespace) carries an order of magnitude more energy than any content role, and structural KEY tokens are over-retained at roughly 1.8x the rate of the answer-carrying VALUE tokens, collapsing exact-match accuracy from 88% to 0% at a 5% budget as the signal-to-noise ratio of the retained state degrades. A counterfactual experiment establishes that suppressing KEY tokens is the best deployable filter. Our retraining-free, role-conditional allocation over SnapKV's windowed score, governed by a single tuned hyperparameter, closes 63-98% of the H2O gap at sub-20% budgets and, at higher budgets, modestly matches or exceeds full-cache accuracy -- a small, seed-sensitive denoising effect (borderline significant at B=0.50; not distinguishable from zero at B=0.30 over four seeds). A 15 MB linear role probe supplies these labels at negligible inference cost, though matching parser-level downstream accuracy remains open.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13205
PDF
-
Text2Sign: A Single-GPU Diffusion Baseline for Text-to-Sign Language Video Generation
2026-07-14 18:15:32Ruize XiaAbstract
Sign language is a primary communication channel for millions of Deaf and hard-of-hearing people, yet text-to-signer video generation remains costly because video diffusion models are expensive to train and evaluate. This paper presents Text2Sign, a text-conditioned diffusion model for short sign-language clips that runs on a single NVIDIA L4 GPU. It combines a frozen vision-language text encoder with a 3D encoder-decoder and factorized spatiotemporal attention to reduce the cost of full-video attention while preserving motion coherence. We compare convolution-only and transformer-style backbones, frozen pretrained and task-specific text encoders, and factorized versus full attention. On a signer-disjoint How2Sign split, the best short-run ablation reaches a validation loss of 0.0648, while a longer-run checkpoint reaches 0.00999. On a compact evaluation slice, the latter achieves an SSIM of $0.2403 \pm 0.0238$, a PSNR of $15.11 \pm 0.42$ dB, and temporal consistency of $1.0000 \pm 0.0000$ using 8-step DDIM sampling with a guidance scale of 5.0. It generates a 32-frame, $64 \times 64$ clip in 12.60 seconds, or 2.54 frames per second, with peak inference memory of 3.12 GB. A held-out denoising audit shows only weak prompt sensitivity: removing text increases late-timestep loss from 0.9875 to 0.9891, while shuffled prompts perform similarly to correct prompts. Frozen text conditioning therefore improves short-budget validation loss, but prompt-specific separation remains limited. The system is restricted to low-resolution, short clips and lacks expert linguistic evaluation, so it should be viewed as a single-GPU research baseline rather than a complete sign-language production system. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13164
PDF
-
The Seriality Gap in Video Diffusion Models
2026-07-14 17:59:22Jorge Diaz Chao, Konpat Preechakul, Yuxi Liu, Yutong BaiAbstract
When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of standard bidirectional video diffusion degrades as the causal chain lengthens, even when provided more denoising steps. In a length-matched single-ball control, where ball-ball interactions are absent, the degradation largely disappears, isolating dependent-event structure rather than video length as the cause. Across intervention studies, methods that increase effective serial computation improve performance disproportionately, including autoregressive/blockwise generation and architectural depth. We identify this pattern as the seriality gap: a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial compute. We then prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13031
PDF
-
Audio-Native Speech Recognition with a Frozen Discrete-Diffusion Language Model
2026-07-14 17:53:22Harsha Vardhan Khurdula, Abhinav Kumar Singh, Yoeven D Khemlani, Vineet AgarwalAbstract
Automatic speech recognition is dominated by autoregressive decoders that emit one token at a time. We ask whether a discrete diffusion language model can transcribe speech instead, refining a whole transcript in parallel over a small number of denoising steps. We train an audio-native interface for DiffusionGemma, a 26B mixture-of-experts model that generates text by uniform, random-token discrete diffusion rather than the absorbing-mask scheme common to recent diffusion language models. A frozen Whisper encoder supplies acoustic features, a lightweight projector maps them into the model embedding space, and low-rank adapters let the frozen backbone attend to the new modality. About 42M parameters are trained, which is 0.16 percent of the backbone. We find that the natural training objectives fail to ground the audio because their gradient reaches the projector only through attention that has already dismissed it. A connectionist temporal classification loss applied through the frozen output head breaks this deadlock. The resulting model reaches 6.6 percent word error rate on LibriSpeech test-clean, transcribes in roughly eight parallel steps regardless of utterance length, and uses a single adapter trained on six languages, which we evaluate here on English, Hindi, and Mandarin.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13013
PDF
-
Low-Latency Neural Models for Real-Time Music Enhancement
2026-07-14 15:22:18Emmanouil Karystinaios, Jonathan Greif, David Nadrchal, Paul Primus, Gerhard WidmerAbstract
Music recordings and live streams are often affected by noise, reverberation, spectral imbalances, or artifacts that degrade listening quality. While speech enhancement has matured into a well-defined research area, music enhancement is less established because musical signals combine overlapping sources, wide bandwidths, strong dynamics, and intentional production effects. We study real-time music enhancement under strict causal and low-latency constraints. We formulate the task around recovery of the intended produced mix from acoustic and production-oriented degradations, adapt compact causal networks to music, and compare speech-derived real-time baselines, an external music-denoising model, an offline restoration reference, and a music-specific MusicFilterNet-MS variant. On the tested hardware, all causal models run faster than real time, but improvements depend strongly on the dataset, degradation type, and metric family; under several objective criteria, indiscriminate enhancement can worsen the degraded input. The main contribution is therefore a benchmark and an analysis rather than a universal best model: real-time music enhancement is feasible, but robust improvement requires degradation-aware modeling, stereo-aware processing, identity-preserving correction, and evaluation beyond a single objective score.
Abstract (translated)
URL
https://arxiv.org/abs/2607.12872
PDF
