Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Real-World Adverse Weather Image Restoration: Enhancing Clearness and Semantics with Vision-Language Models

Jiaqi Xu, Mengyang Wu, Xiaowei Hu, Chi-Wing Fu, Qi Dou, Pheng-Ann Heng

Abstract

This paper addresses the limitations of adverse weather image restoration approaches trained on synthetic data when applied to real-world scenarios. We formulate a semi-supervised learning framework employing vision-language models to enhance restoration performance across diverse adverse weather conditions in real-world settings. Our approach involves assessing image clearness and providing semantics using vision-language models on real data, serving as supervision signals for training restoration models. For clearness enhancement, we use real-world data, utilizing a dual-step strategy with pseudo-labels assessed by vision-language models and weather prompt learning. For semantic enhancement, we integrate real-world data by adjusting weather conditions in vision-language model descriptions while preserving semantic meaning. Additionally, we introduce an effective training strategy to bootstrap restoration performance. Our approach achieves superior results in real-world adverse weather image restoration, demonstrated through qualitative and quantitative comparisons with state-of-the-art works.

Towards Real-World Adverse Weather Image Restoration: Enhancing Clearness and Semantics with Vision-Language Models

Abstract

This paper addresses the limitations of adverse weather image restoration approaches trained on synthetic data when applied to real-world scenarios. We formulate a semi-supervised learning framework employing vision-language models to enhance restoration performance across diverse adverse weather conditions in real-world settings. Our approach involves assessing image clearness and providing semantics using vision-language models on real data, serving as supervision signals for training restoration models. For clearness enhancement, we use real-world data, utilizing a dual-step strategy with pseudo-labels assessed by vision-language models and weather prompt learning. For semantic enhancement, we integrate real-world data by adjusting weather conditions in vision-language model descriptions while preserving semantic meaning. Additionally, we introduce an effective training strategy to bootstrap restoration performance. Our approach achieves superior results in real-world adverse weather image restoration, demonstrated through qualitative and quantitative comparisons with state-of-the-art works.
Paper Structure (31 sections, 6 equations, 10 figures, 2 tables)

This paper contains 31 sections, 6 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Image Restoration in Adverse Weather Conditions
  4. Vision-Language Models
  5. Methodology
  6. Enhancing Image Clearness through Vision-Language Models
  7. Image Assessment and Pseudo-Labeling.
  8. Image assessment.
  9. Pseudo-labeling.
  10. Weather Prompt Learning.
  11. Prompt embedding learning.
  12. Restoration model optimization.
  13. Enhancing Image Semantics through Vision-Language Models
  14. Description-assisted semantic enhancement.
  15. Loss function.
  16. ...and 16 more sections

Figures (10)

  • Figure 1: The clearness level and the semantics information of real-world adverse weather images are provided by large vision-language models. This assistance is instrumental in training image restoration models to effectively utilize real-world data.
  • Figure 2: The schematic illustration of the proposed semi-supervised learning framework enhancing real-world image restoration by improving clearness and semantics in varied adverse weather conditions through the utilization of vision-language models.
  • Figure 3: VLM-based assessment of images restored from weather-related artifacts. In (a), we show the process to compute the VLM's image assessment ratings $r^{vlm}$. In (b), we find that $r^{vlm}$ can select pseudo-labels with fewer weather-related artifacts.
  • Figure 4: The workflow of the weather prompt learning approach.
  • Figure 5: Description-assisted semantic enhancement with the vision-language models.
  • ...and 5 more figures