DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better

We present a new end-to-end generative adversarial network (GAN) for singleimage motion deblurring, named DeblurGAN-v2, which considerably boostsstate-of-the-art deblurring efficiency, quality, and flexibility. DeblurGAN-v2is based on a relativistic conditional GAN with a double-scale discriminator.For the first time, we introduce the Feature Pyramid Network into deblurring,as a core building block in the generator of DeblurGAN-v2. It can flexibly workwith a wide range of backbones, to navigate the balance between performance andefficiency. The plug-in of sophisticated backbones (e.g., Inception-ResNet-v2)can lead to solid state-of-the-art deblurring. Meanwhile, with light-weightbackbones (e.g., MobileNet and its variants), DeblurGAN-v2 reaches 10-100 timesfaster than the nearest competitors, while maintaining close tostate-of-the-art results, implying the option of real-time video deblurring. Wedemonstrate that DeblurGAN-v2 obtains very competitive performance on severalpopular benchmarks, in terms of deblurring quality (both objective andsubjective), as well as efficiency. Besides, we show the architecture to beeffective for general image restoration tasks too. Our codes, models and dataare available at: https://github.com/KupynOrest/DeblurGANv2