DeblurGAN-v2 is a novel end-to-end generative adversarial network (GAN) designed for single-image motion deblurring, aiming to significantly enhance the efficiency, quality, and flexibility of existing methods. The model introduces the Feature Pyramid Network (FPN) as a core component in the generator, allowing it to work flexibly with various backbones, balancing performance and efficiency. DeblurGAN-v2 uses a relativistic conditional GAN with a double-scale discriminator, which evaluates both global and local scales. The model achieves state-of-the-art performance on popular benchmarks, including the GoPro, DVD, and NFS datasets, with significant improvements in both PSNR, SSIM, and perceptual quality. Notably, DeblurGAN-v2 with lightweight backbones like MobileNet and MobileNet-DSC is 10-100 times faster than its nearest competitors while maintaining near-state-of-the-art results, making it suitable for real-time video deblurring. The paper also demonstrates the model's effectiveness in general image restoration tasks and provides detailed experimental evaluations and ablation studies to support its claims.DeblurGAN-v2 is a novel end-to-end generative adversarial network (GAN) designed for single-image motion deblurring, aiming to significantly enhance the efficiency, quality, and flexibility of existing methods. The model introduces the Feature Pyramid Network (FPN) as a core component in the generator, allowing it to work flexibly with various backbones, balancing performance and efficiency. DeblurGAN-v2 uses a relativistic conditional GAN with a double-scale discriminator, which evaluates both global and local scales. The model achieves state-of-the-art performance on popular benchmarks, including the GoPro, DVD, and NFS datasets, with significant improvements in both PSNR, SSIM, and perceptual quality. Notably, DeblurGAN-v2 with lightweight backbones like MobileNet and MobileNet-DSC is 10-100 times faster than its nearest competitors while maintaining near-state-of-the-art results, making it suitable for real-time video deblurring. The paper also demonstrates the model's effectiveness in general image restoration tasks and provides detailed experimental evaluations and ablation studies to support its claims.