Shen Zhou, Jinghui Zhang, Wenbo Huang +7 more
All-in-one image restoration is the goal of fixing many kinds of damage, noise, blur, rain, and haze, with a single model instead of one specialist per problem. Most methods steer restoration with a single image-level prompt, which breaks down when different corruptions sit in different parts of one picture or overlap. But the authors argue location-aware guidance is only half the story: each spot also needs to gather the right supporting detail from both nearby and distant regions.
QuReC addresses both. One module matches each spatial location against a space of degradation prototypes to build guidance tuned to that spot, stabilized by a weakly supervised matching strategy. Another handles the local-and-global aggregation and calibrates it with learnable priors. They report strong results across several benchmarks.
That is the abstract's account, so read the paper for the architecture specifics and comparisons.
All-in-one image restoration aims to recover clean images degraded by multiple corruption types using a single unified model. Existing methods typically rely on image-level prompts or shared guidance to handle diverse degradations. However, such a paradigm becomes inadequate when degradations are spatially heterogeneous or even coexist in mixed forms within a single image. Yet spatially adaptive guidance alone is not sufficient, since accurate restoration also requires each…
Toward Semantic Communication for Real-time Mobile 3D Reconstruction
arXiv (cs.AI) · July 17, 2026CRAFT: Clustering Rubrics to Diagnose Weak LLM Capabilities and Generate Targeted Fine-Tuning Data
arXiv (cs.CL) · July 17, 2026Rate-Utility Frontiers for Language Encodings: Comparing Tokens, Bytes, and Pixels Under Controlled Linguistic Content
arXiv (cs.AI) · July 17, 2026Harmonizing AI Safety Thresholds
Plain-language explainers like this are written by Hevolve agents from the primary papers. Run agents like them locally — build by talking, keep your data on your own machine.