Real-World Super-Resolution via Kernel Estimation and Noise Injection
Recent state-of-the-art super-resolution methods have achieved impressive performance on ideal datasets regardless of blur and noise. However, these methods always fail in real-world image super-resolution, since most of them adopt simple bicubic downsampling from high-quality images to construct Low-Resolution (LR) and High-Resolution (HR) pairs for training which may lose track of frequency-related details. To address this issue, we focus on designing a novel degradation framework for real- world images by estimating various blur kernels as well as real noise distributions. Based on our novel degradation framework, we can acquire LR images sharing a common domain with real-world images. Then, we propose a real- world super-resolution model aiming at better perception. Extensive experiments on synthetic noise data and real- world images demonstrate that our method outperforms the state-of-the-art methods, resulting in lower noise and better visual quality. In addition, our method is the winner of NTIRE 2020 Challenge on both tracks of Real-World Super-Resolution, which significantly outperforms other competitors by large margins.
PDFCode
MSU Video Super Resolution Benchmark: Detail Restoration
MSU Video Upscalers: Quality Enhancement
MSU Super-Resolution for Video Compression
Results from the Paper
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Video Super-Resolution | MSU Super-Resolution for Video Compression | RealSR + aomenc | BSQ-rate over ERQA | 6.762 | # 33 | |
| BSQ-rate over Subjective Score | 0.843 | # 15 | ||||
| BSQ-rate over VMAF | 4.283 | # 59 | ||||
| BSQ-rate over PSNR | 15.144 | # 73 | ||||
| BSQ-rate over MS-SSIM | 5.463 | # 62 | ||||
| BSQ-rate over LPIPS | 10.915 | # 55 | ||||
| Video Super-Resolution | MSU Super-Resolution for Video Compression | RealSR + uavs3e | BSQ-rate over ERQA | 1.943 | # 16 | |
| BSQ-rate over Subjective Score | 0.639 | # 8 | ||||
| BSQ-rate over VMAF | 2.253 | # 50 | ||||
| BSQ-rate over PSNR | 14.741 | # 72 | ||||
| BSQ-rate over MS-SSIM | 1.441 | # 28 | ||||
| BSQ-rate over LPIPS | 1.149 | # 8 | ||||
| Video Super-Resolution | MSU Super-Resolution for Video Compression | RealSR + x265 | BSQ-rate over ERQA | 1.622 | # 11 | |
| BSQ-rate over Subjective Score | 0.502 | # 7 | ||||
| BSQ-rate over VMAF | 1.617 | # 36 | ||||
| BSQ-rate over PSNR | 1.064 | # 5 | ||||
| BSQ-rate over MS-SSIM | 1.033 | # 22 | ||||
| BSQ-rate over LPIPS | 1.206 | # 9 | ||||
| Video Super-Resolution | MSU Super-Resolution for Video Compression | RealSR + vvenc | BSQ-rate over ERQA | 21.965 | # 82 | |
| BSQ-rate over VMAF | 10.67 | # 81 | ||||
| BSQ-rate over PSNR | 15.144 | # 73 | ||||
| BSQ-rate over MS-SSIM | 11.643 | # 82 | ||||
| BSQ-rate over LPIPS | 18.344 | # 85 | ||||
| Video Super-Resolution | MSU Super-Resolution for Video Compression | RealSR + x264 | BSQ-rate over ERQA | 0.77 | # 2 | |
| BSQ-rate over Subjective Score | 0.196 | # 1 | ||||
| BSQ-rate over VMAF | 0.775 | # 14 | ||||
| BSQ-rate over PSNR | 0.675 | # 1 | ||||
| BSQ-rate over MS-SSIM | 0.487 | # 1 | ||||
| BSQ-rate over LPIPS | 0.591 | # 2 | ||||
| Video Super-Resolution | MSU Video Super Resolution Benchmark: Detail Restoration | RealSR | Subjective score | 5.286 | # 17 | |
| ERQAv1.0 | 0.69 | # 12 | ||||
| QRCRv1.0 | 0 | # 21 | ||||
| SSIM | 0.767 | # 30 | ||||
| PSNR | 25.989 | # 21 | ||||
| FPS | 0.352 | # 25 | ||||
| 1 - LPIPS | 0.911 | # 8 | ||||
| Video Super-Resolution | MSU Video Upscalers: Quality Enhancement | RealSR | PSNR | 30.64 | # 14 | |
| LPIPS | 0.220 | # 15 | ||||
| SSIM | 0.900 | # 33 |
