Image Harmonization and Compositing Quiz

Answers and explanations appear as you take the quiz.

1. What problem does image harmonization primarily solve?
   • Matching a pasted object's color and lighting to its new background
   • Removing image noise
   • Increasing image resolution
   • Detecting faces in photos
2. What extra input do most harmonization models need besides the composite image?
   • A text caption
   • A binary mask of the inserted region
   • The camera's GPS data
   • An audio clip
3. What is the iHarmony4 benchmark associated with?
   • Speech recognition
   • Self-driving cars
   • Image harmonization research
   • Music generation
4. How are training pairs for harmonization commonly created?
   • By manually painting every image
   • By using only synthetic 3D renders
   • By recording video
   • By disturbing the colors of a region in a real photo to create a fake composite
5. Why do many efficient methods predict a color transform instead of regenerating pixels?
   • To preserve the foreground's detail and texture
   • To use more memory
   • To make images smaller
   • To avoid using a mask
6. What capability are newer diffusion-based compositing models adding beyond color matching?
   • Translating text languages
   • Synthesizing matching shadows and reflections
   • Compressing video files
   • Recognizing handwriting
7. What does DoveNet's domain-verification idea ask a harmonized foreground to do?
   • Become a separate class that is easier to detect
   • Replace the background with the foreground
   • Move toward the same appearance domain as the background
   • Ignore the background's lighting and color
8. Which source datasets contribute to iHarmony4's harmonization composites?
   • ImageNet, LAION-5B, and Common Crawl subsets
   • Only synthetic renderings from Objaverse
   • Only unedited satellite image collections
   • COCO, Adobe5k, Flickr, and day2night subsets