Real-World Video Datasets for Haze Removal

Description for the Video Dehazing Dataset

The proposed dataset comprises 22 synthetic hazy videos, each carefully designed to simulate varying levels of haze intensity and diverse environmental conditions, such as different lighting, weather patterns, and scene complexities. Each video was converted into individual frames, resulting in a comprehensive dataset suitable for training and evaluating advanced video dehazing algorithms.

This dataset was utilized to explore a novel machine-learning-based approach that leverages the UNet architecture in conjunction with a linear variance scheduler within the diffusion process framework. The frames serve as input to the dehazing model, enabling the system to learn spatiotemporal features effectively.

The dataset provides a valuable benchmark for researchers focusing on video dehazing and restoration tasks, offering high-quality synthesized data to test innovative techniques in image enhancement and haze removal. It is particularly suited for algorithms requiring extensive frame-by-frame processing while preserving temporal consistency.