100 natural images representing the testing subset of the Berkeley Segmentation Dataset (BSD300). These images are used as a standard benchmark for evaluating image restoration tasks such as super-resolution and denoising.
Use Cases
- Benchmark single-image super-resolution (SISR) models by comparing reconstructed outputs against these 100 ground-truth images
- Test image denoising algorithms by applying synthetic noise and measuring recovery quality
- Evaluate edge detection performance using the structural boundaries inherent in the natural scenes
Strengths
- 100 natural images in the testing partition
- Derived from the BSD300 Berkeley Segmentation Dataset
- Standard benchmark for PSNR and SSIM metric evaluation in image restoration