GAN-Based Image Dehazing on Raspberry PI

Dehazing is a critical step in computer vision that seeks to remove atmospheric haze from images. In recent times, Generative Adversarial Networks (GANs) have gained significant attention as a powerful dehazing and image restoration solution. This study proposes a GAN-based dehazing model specifically designed for atmospheric haze removal in images. The model incorporates generator and discriminator components to enhance the dehazing process through adversarial training. To effectively capture spatial and contextual information, various architectures such as MANet, PSPNet, and FPN are explored. Additionally, the model incorporates a Vision Transformer as an encoder block to improve feature extraction. Using objective measures like PSNR and SSIM, the proposed technique is assessed and trained on the RESIDE dataset. Furthermore, the model is implemented for real-time dehazing on a Raspberry Pi device. The results of the experiments and comparative studies offer convincing proof that the suggested strategy is successful in producing high-quality dehazing results.