Orest Kupyn, Volodymyr Budzan, Mykola Mykhailych, Dmytro Mishkin, Jiri Matas
We present DeblurGAN, an end-to-end learned method for motion deblurring. The learning is based on a conditional GAN and the content loss . DeblurGAN achieves state-of-the art performance both in the structural similarity measure and visual appearance. The quality of the deblurring model is also evaluated in a novel way on a real-world problem -- object detection on (de-)blurred images. The method is 5 times faster than the closest competitor -- DeepDeblur. We also introduce a novel method for generating synthetic motion blurred images from sharp ones, allowing realistic dataset augmentation. The model, code and the dataset are available at https://github.com/KupynOrest/DeblurGAN
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Deblurring | REDS | Average PSNR | 24.09 | DeblurGAN |
| Deblurring | RealBlur-R (trained on GoPro) | SSIM (sRGB) | 0.903 | DeblurGAN |
| Deblurring | RealBlur-J (trained on GoPro) | SSIM (sRGB) | 0.834 | DeblurGAN |
| 2D Classification | REDS | Average PSNR | 24.09 | DeblurGAN |
| 2D Classification | RealBlur-R (trained on GoPro) | SSIM (sRGB) | 0.903 | DeblurGAN |
| 2D Classification | RealBlur-J (trained on GoPro) | SSIM (sRGB) | 0.834 | DeblurGAN |
| 10-shot image generation | REDS | Average PSNR | 24.09 | DeblurGAN |
| 10-shot image generation | RealBlur-R (trained on GoPro) | SSIM (sRGB) | 0.903 | DeblurGAN |
| 10-shot image generation | RealBlur-J (trained on GoPro) | SSIM (sRGB) | 0.834 | DeblurGAN |
| Blind Image Deblurring | REDS | Average PSNR | 24.09 | DeblurGAN |
| Blind Image Deblurring | RealBlur-R (trained on GoPro) | SSIM (sRGB) | 0.903 | DeblurGAN |
| Blind Image Deblurring | RealBlur-J (trained on GoPro) | SSIM (sRGB) | 0.834 | DeblurGAN |