Image Super-Resolution Quiz

Answers and explanations appear as you take the quiz.

1. Why is single-image super-resolution called an 'ill-posed' problem?
   • Many different high-resolution images could produce the same low-resolution input
   • The input images are always corrupted beyond repair
   • The math has no solution at all
   • It requires more memory than any GPU has
2. What is a common drawback of training super-resolution networks with only a pixel-wise MSE loss?
   • Outputs become overly sharp and noisy
   • Outputs look blurry and over-smoothed
   • The model refuses to enlarge the image
   • It can only process grayscale images
3. What does the GAN-based model ESRGAN add to improve perceived sharpness?
   • Larger bicubic interpolation kernels
   • A rule-based edge sharpening filter
   • A discriminator and perceptual feature-space loss
   • More training epochs only
4. Why must super-resolution outputs be treated cautiously in forensic or medical settings?
   • They are always lower quality than the input
   • They cannot be saved in standard formats
   • They double the file size every time
   • The model invents plausible detail that may not be the true content
5. How does diffusion-based super-resolution (like SR3) generate detail?
   • By iteratively refining noise into detail over many steps
   • By copying pixels from a database of stock photos
   • By applying a single sharpening convolution
   • By compressing the image with JPEG
6. What problem does Real-ESRGAN specifically target compared with earlier SR models?
   • Upscaling only cartoon images
   • Handling messy real-world degradations like noise and compression
   • Reducing image resolution on purpose
   • Converting images to vector graphics
7. What does the sub-pixel convolution in ESPCN do?
   • It estimates optical flow between video frames
   • It classifies objects without upsampling
   • It rearranges feature channels into a higher-resolution image grid
   • It adds a text prompt to an image embedding
8. What kind of image pairs did the RealSR dataset collect?
   • Only synthetic Gaussian-noise images
   • Text-only descriptions paired with audio
   • Unpaired medical scans from different patients
   • Real-world paired low- and high-resolution images captured with different cameras or zoom settings