Enhancing degradation robustness is essential for deploying image fusion techniques in real-world dynamic scenes. However, most existing methods either handle a single degradation type or assume fixed multi-degradation settings, making them insufficient for dynamically heterogeneous and composition ally complex degradations in practice. Moreover, they often fail to recover the semantics of salient scene targets when these targets are degraded or missing, leading to weakened semantic representation and reduced target saliency. To address these challenges, we propose DuS-DiFuse, a robust dual-stream latent diffusion framework composed of a diffusion fusion unit and a generative modulation unit. In the diffusion fusion unit, we fine-tune a CLIP visual encoder on multi-source data to perceive degradation types and severities, and employ latent diffusion to uniformly model multi-type, cross-level degradations with varying parameters. A Groupwise Fusion Control Module (GFCM) is further embedded into the latent degradation-removal process, enabling joint modeling of dynamic degradation removal and multimodal information fusion. In the generative modulation unit, pretrained latent diffusion priors are used to remodulate the initial fusion results, enabling controllable semantic restoration and generative enhancement, thereby improving target saliency and overall visual quality. To preserve fine-grained details during latent-to image reconstruction, we introduce a Detail-Restoration Fidelity Module (DRFM), which constrains texture reconstruction by jointly leveraging multi-level skip features from multiple source images and enhances structural fidelity in the fused results. Extensive experiments on multiple fusion datasets demonstrate that DuS-DiFuse achieves leading fusion performance, exhibits strong robustness to heterogeneous degradations, generalizes well across fusion tasks, and supports effective controllable generative modulation.
Lei Cao, Hao Zhang, Peng Zhang et al.· IEEE Transactions on Pattern...· 0 citations
Instruction-based general video editing seeks to unify diverse editing operations within a single, intuitive interface. Existing approaches often rely on resource-intensive conditioning, using either heavyweight branches or costly source concatenation. Is there any efficient way to model editing intent? Thus, we introduce GRNEdit, a lightweight two-stage framework. GRN inspires our approach by encoding visual semantics through combinations of bits. Through task-specific fine-tuning, we take this representation further and recast editing semantics as local retain-or-flip decisions over individual bits. Source information is consequently modeled as coordinate-wise evidence supporting the observed binary states, while the GRN backbone remains responsible for resolving their global composition into coherent generative semantics. In Stage I, a compact encoder translates discrete source codes into continuous evidence signals, which GRN assimilates throughout binary refinement. Inspired by null-prompt training for classifier-free guidance, we further assign the null condition an editing-specific meaning: an empty instruction denotes no edit and is supervised through source reconstruction. This identity pathway not only implicitly strengthens evidence utilization and content preservation in Stage I, but also produces a source-preserving state in the same representation space as the edited state. Stage II can therefore directly compare each edited state with its source-preserving counterpart and use their discrepancy to revise unresolved target-bit decisions. Trained on only 0.6M pairs with less than 3\% conditioning parameters, GRNEdit-2B and GRNEdit-8B achieve scores of 4.03 and 4.18 on OpenVE-Bench. The 2B model outperforms multiple 14B open-source editors, while the 8B model performs on par with leading open-source editors.
Streaming autoregressive video models generate long videos chunk by chunk, using historical memory to maintain consistency. Existing methods typically expose subject and scene queries to history through similar policies. This stabilizes the subject, but can also lock backgrounds, viewpoints, and scene structure to previously generated states even when local motion continues. We call this failure memory-anchored scene under-progression; consistency and motion metrics alone can miss it. We introduce TetherMem, a training-free, query-aware spatiotemporal memory router for frozen video generators. TetherMem separates subject and scene queries and modulates historical access with region- and age-conditioned priors: subject queries retain identity-bearing history, while scene queries reduce reliance on subject history and stale backgrounds. Across 2,400 blinded pairwise judgments from 10 annotators, TetherMem achieves the highest estimated expected preference among eight streaming long-video baselines for overall quality (0.780) and scene progression (0.769). On complete 30-second videos, it sustains changes in background, viewpoint, and scene state while preserving subject recognizability and temporal continuity.
Chen Li, Peng Zhang, Hanyu Zhou et al.· 0 citations
