Prompt-based All-in-One Image Restoration Network For Robust Visual Odometry

Conventional image restoration models are difficult to apply efficiently in real-world scenarios because they are designed to handle only specific types and levels of degradation. This study proposes a model that can handle multiple degradations with a single restoration model using prompt learning. Furthermore, we introduce a sub-network, pixel-level encoder, that modulates the encoder of the main network and prompts. In this process, the proposed model adaptively integrates features across spatial and channel spaces through deformable spatial cross-attention and multi-Dconv head transposed cross-attention. Moreover, this model exploits pixel-wise contrastive loss to capture the style and context information of target images. An experimental evaluation is conducted using a widely-used dataset in all-in-one image restoration, including dehazing, deraining, and denoising. Additionally, this study evaluates the robustness of models for monocular depth estimation and visual odometry using images reconstructed from noise degradation.