Generalizing to Out-of-Sample Degradations via Model Reprogramming

Existing image restoration models are typically designed for specific tasks and struggle to generalize to out-of-sample degradations not encountered during training. While zero-shot methods can address this limitation by fine-tuning model parameters on testing samples, their effectiveness relies on predefined natural priors and physical models of specific degradations. Nevertheless, determining out-of-sample degradations faced in real-world scenarios is always impractical. As a result, it is more desirable to train restoration models with inherent generalization ability. To this end, this work introduces the Out-of-Sample Restoration (OSR) task, which aims to develop restoration models capable of handling out-of-sample degradations. An intuitive solution involves pre-translating out-of-sample degradations to known degradations of restoration models. However, directly translating them in the image space could lead to complex image translation issues. To address this issue, we propose a model reprogramming framework, which translates out-of-sample degradations by quantum mechanic and wave functions. Specifically, input images are decoupled as wave functions of amplitude and phase terms. The translation of out-of-sample degradation is performed by adapting the phase term. Meanwhile, the image content is maintained and enhanced in the amplitude term. By taking these two terms as inputs, restoration models are able to handle out-of-sample degradations without fine-tuning. Through extensive experiments across multiple evaluation cases, we demonstrate the effectiveness and flexibility of our proposed framework. Our codes are available at https://github.com/ddghjikle/Out-of-sample-restoration.

I. Introduction

Image restoration have achieved remarkable success with the rapid development of deep neural networks. Previous researches [1], [2], [3] always characterize specific types of degradations as individual issues and propose dedicated solutions. While this methodology effectively addresses some real-world scenarios, it falls short when considering more complex situations like autonomous driving on rainy days. In such cases, the perceived images can be degraded by a combination of rain, haze, blur, and noise, making it difficult to attribute the degradation to a specific form. Moreover, as real-world images can exhibit diverse and unpredictable degradation patterns, enumerating all possible degradations to train restoration networks is practically infeasible [4]. Therefore, it is imperative to develop restoration models that can effectively restore images degraded by various factors, including those not encountered during training.