RESIDE-6K

The RESIDE-6K dataset is a subset of the RESIDE (REalistic Single Image DEhazing) dataset, specifically organized by the creators of DehazeFormer. It provides an excellent benchmark for researchers and developers focusing on image dehazing algorithms.

Dataset Overview:

• Image Pairs: Each hazy image is paired with its corresponding haze-free ground truth.
• Training Data: 6,000 image pairs, including:
  • Indoor Training Set (ITS): 3,000 image pairs.
  • Outdoor Training Set (OTS): 3,000 image pairs.
  • All images are resized to 400×400 pixels.
• Testing Data: 1,000 image pairs, equally split:
  • 500 indoor scenes.
  • 500 outdoor scenes.
  • No resizing applied to testing images.

Key Features:

1. High-Quality Pairs: Accurate alignment of hazy and ground truth images for precise dehazing model evaluation.
2. Diverse Scenarios: Balanced dataset with both indoor and outdoor environments for a comprehensive analysis.
3. Standardized Dimensions: Training images resized to 400×400 pixels for consistency.

Applications:

• Training and benchmarking image dehazing models.
• Enhancing visibility in computer vision tasks like object detection, scene segmentation, and surveillance.
• Improving real-world applications in photography, autonomous driving, and aerial imagery.

This dataset is sourced from Kaggle.