Efficient Degradation-aware Any Image Restoration

Reconstructing missing details from degraded low-quality inputs poses asignificant challenge. Recent progress in image restoration has demonstratedthe efficacy of learning large models capable of addressing variousdegradations simultaneously. Nonetheless, these approaches introduceconsiderable computational overhead and complex learning paradigms, limitingtheir practical utility. In response, we propose \textit{DaAIR}, an efficientAll-in-One image restorer employing a Degradation-aware Learner (DaLe) in thelow-rank regime to collaboratively mine shared aspects and subtle nuancesacross diverse degradations, generating a degradation-aware embedding. Bydynamically allocating model capacity to input degradations, we realize anefficient restorer integrating holistic and specific learning within a unifiedmodel. Furthermore, DaAIR introduces a cost-efficient parameter updatemechanism that enhances degradation awareness while maintaining computationalefficiency. Extensive comparisons across five image degradations demonstratethat our DaAIR outperforms both state-of-the-art All-in-One models anddegradation-specific counterparts, affirming our efficacy and practicality. Thesource will be publicly made available at https://eduardzamfir.github.io/daair/