🚗 Clear Roads, Clear Vision

🌦️ A Comprehensive Review on Multi-Weather Restoration for Smart Transportation

Welcome to the official repository accompanying our survey paper:

“Clear Roads, Clear Vision: Advancements in Multi-Weather Restoration for Smart Transportation” 📌 Submitted to IEEE Transactions on Intelligent Transportation Systems, 2025

📘 About This Repository

This repository presents a curated collection of literature referenced in our comprehensive survey on image and video restoration methods for adverse weather conditions relevant to intelligent transportation. These methods tackle degradation caused by haze, rain, snow, or a combination of them using deep learning and classical techniques.

📚 Categorized Restoration Literature

🌫 Haze Removal — Removing fog and haze to enhance visibility

| Paper Title | Link | | ------------------------------------------------------------------------------------------ | ------------------------------------------------------------- | | DehazeNet: An End-to-End System for Single Image Haze Removal | Link | | AOD-Net: All-in-One Dehazing Network | Link | | Single Image Dehazing Using Haze-Lines | Link | | Single Image Haze Removal Using Dark Channel Prior | Link | | FFA-Net: Feature Fusion Attention Network for Single Image Dehazing | Link | | Visibility in Bad Weather from a Single Image | Link | | A Fast Single Image Dehazing Algorithm Based on Artificial Multiexposure Image Fusion | Link | | Restormer: Efficient Transformer for High-Resolution Image Restoration | Link | | FCANet: Frequency Channel Attention for Image Dehazing | Link | | FD-GAN: Generative Adversarial Network With Fusion-Discriminator for Single Image Dehazing | Link |

🌧 Rain Removal — Tackling streaks and accumulation from rainfall

| Paper Title | Link | | ----------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Deep Joint Rain Detection and Removal from a Single Image | Link | | Removing Rain from Single Images via a Deep Detail Network | Link | | Density-aware Image Deraining using a Multi-stream Dense Network | Link | | Recurrent Squeeze-and-Excitation Context Aggregation Net for Single Image Deraining | Link | | Spatial Attentive Single-Image Deraining with a High Quality Real Rain Dataset | Link | | Multi-Scale Progressive Fusion Network for Single Image Deraining | Link | | Unsupervised Single Image Deraining with Rain Guidance Feature | Link | | Learning a Smart Knowledge Assignment Strategy for Deraining and Desnowing | Link | | Deraining via Dual Graph Convolutional Network | Link | | Residual Dense Network for Image Denoising | Link |

❄️ Snow Removal — Eliminating snowy occlusions from scenes

| Paper Title | Link | | ---------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | | DesnowNet: Context-Aware Deep Network for Snow Removal | Link | | DesnowGAN: Snow Removal via GAN-Based Residual Estimation | Link | | JSTASR: Joint Snow Removal and Scene Restoration Using Attention Mechanisms | Link | | DesnowFormer: Transformer-Based Snow Removal via Multi-Scale and Residual Guidance | Link | | MSP-Former: Multi-Stage Progressive Transformer for Single Image Desnowing | Link | | SnowFormer: A Transformer-Based Framework for Snow Removal | Link | | Deep Invertible Separation for Snow Removal | Link | | FPGA-based Lightweight Snow Removal | Link | | Marine Snow Removal in Underwater Scenes | Link |

🧠 All-in-One Restoration — Unified solutions across multiple degradations

Prompt-Guided

| Paper Title | Link | | ----------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | PromptIR: Prompting for All-in-One Image Restoration | Link | | Language-Driven All-in-One Restoration | Link | | DPMambaIR: Degradation-Aware Prompt State Space Model | Link |

Transformer-Based

| Paper Title | Link | | --------------------------------------------------------------------------------- | ----------------------------------------------- | | TransWeather: Transformer-based Restoration of Images Degraded by Adverse Weather | Link | | GridFormer: Grid-Aware Feature Alignment Transformer for Image Restoration | Link | | Frequency Transformer: Frequency-Aware Degradation Restoration Network | Link |

Diffusion-Based

| Paper Title | Link | | --------------------------------------------------------------------- | ---------------------------------------- | | AutoDIR: Automatic All-in-One Image Restoration with Latent Diffusion | Link | | Visual-Instructed Degradation Diffusion | Link |

Others

| Paper Title | Link | | --------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- | | WM-MoE: Weather-Aware Multi-scale Mixture-of-Experts | Link | | MAC-GAN: Multi-Adversarial Contextual GAN for All-Weather Restoration | Link | | NAS-AIOIR: All in One Bad Weather Removal Using Architectural Search | Link |

☂️ Multi-Weather Restoration — Addressing combined and real-world degradations

Image

| Paper Title | Link | | ------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | | WEAFU: Weather Adaptive Feature Update Network | Link | | MWFormer: Multi-weather Transformer | Link | | Unified Transformer for Multi-degradation Removal | Link | | Gated Context Aggregation Network | Link | | General/Specific Weather Restoration Framework | Link |

Video

| Paper Title | Link | | ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------ | | Video Restoration via Matrix Decomposition | Link | | CANet: Context Aggregation Network | Link | | Dual Spatio-Temporal Transformer Network | Link | | Meta-Adaptation Framework | Link |

✍️ Citation

bibtex @article{your2025paper, title = {Clear Roads, Clear Vision: Advancements in Multi-Weather Restoration for Smart Transportation}, author = {Your Name and Co-authors}, journal = {IEEE Transactions on Intelligent Transportation Systems}, year = {2025}, doi = {YOUR_DOI_HERE} }

📈 Datasets & Benchmarks

This section lists the commonly used datasets referenced in our survey for training and evaluating multi-weather restoration models.

🌫 Haze Datasets

• RESIDE: A benchmark dataset for single image dehazing with synthetic and real-world haze images. Link
• I-HAZE & O-HAZE: Real hazy images captured under controlled settings. Link

🌧 Rain Datasets

• Rain100L / Rain100H: Synthetic datasets containing light and heavy rain streaks. Link
• DID-Data: Dataset used for training DID-MDN, simulating various rain densities. Link

❄️ Snow Datasets

• Snow100K: A large-scale dataset with synthesized snowy images. Link

🧠 All-in-One and Multi-Weather Datasets

• DAWN: Diverse Adverse Weather Needs dataset that includes haze, rain, and snow in a unified benchmark. Link
• Realistic Multi-weather Dataset (RMWD): Contains videos and images under diverse real-world weather.

🎞️ Video Datasets

• REVIDE: Real-world video dehazing dataset. Link

• NTIRE Video Restoration Challenges: Benchmarks for rain, snow, and haze video sequences. Link

• 📊 The repository groups restoration techniques by weather condition to match the structure of the original paper's benchmarking tables.

• 📁 Each method is listed with its title and official source link (DOI, CVF, or arXiv).

• 📈 Representative datasets for each degradation type are included for reproducibility and further research.

• 📌 For full quantitative benchmarks and evaluations, refer to Tables I–V in the original paper.

🤝 How to Contribute

Pull requests are welcome for:

• Adding newer restoration papers
• Providing paper links (arXiv/DOI)
• Improving benchmarks/tables
• Fixing typos or reclassifications