The field of image restoration is experiencing a significant shift towards the development of more efficient, adaptive, and intelligent solutions. Recent research has focused on improving the robustness of deep learning models for extreme image deblurring, leveraging vision-language guidance for universal restoration, and enhancing the scalability of generative models for high-resolution image restoration. Noteworthy papers in this area include X-DECODE, which introduces a novel training strategy based on curriculum learning, and VL-UR, which proposes a vision-language-guided universal restoration framework. Other notable contributions include the development of efficient and mixed heterogeneous models, such as RestorMixer, and the introduction of adaptive quality prompting mechanisms, like AdaQual-Diff. These advancements have the potential to improve the quality and efficiency of image restoration in various applications, including autonomous driving, medical imaging, and photography.
Advances in Image Restoration
Sources
X-DECODE: EXtreme Deblurring with Curriculum Optimization and Domain Equalization
VL-UR: Vision-Language-guided Universal Restoration of Images Degraded by Adverse Weather Conditions
ZipIR: Latent Pyramid Diffusion Transformer for High-Resolution Image Restoration
Enhancing Image Restoration through Learning Context-Rich and Detail-Accurate Features
An Efficient and Mixed Heterogeneous Model for Image Restoration
AdaQual-Diff: Diffusion-Based Image Restoration via Adaptive Quality Prompting
High-Fidelity Image Inpainting with Multimodal Guided GAN Inversion
TTRD3: Texture Transfer Residual Denoising Dual Diffusion Model for Remote Sensing Image Super-Resolution