合成图像检测测验

当您参加测验时，就会出现答案和解释。

1. A detector marks a photo as probably synthetic. Which conclusion is outside that score alone?
   • Whether the image’s depicted claim is true
   • That a model produced a score
   • That further investigation is possible
   • That errors can occur
2. Why test on screenshots and images from newer generators?
   • Any detector always generalizes to new generators
   • Screenshots prove an image is genuine
   • Compression cannot change model features
   • Their artifacts may differ from the detector’s training data
3. A raw detector score is 0.9. What is needed before reading it as a 90% probability?
   • A more colorful UI
   • Calibration on relevant held-out data and threshold context
   • A JPEG extension
   • A missing content credential
4. In the guide’s invented 1,010-image example, 8 synthetic and 10 genuine images are flagged. What fraction of flags are synthetic?
   • Eight of ten
   • Ten of eighteen
   • Eight of eighteen
   • One percent exactly
5. An image lacks Content Credentials. What follows?
   • It must be AI-generated
   • It must be an unedited camera original
   • It has no source history anywhere
   • Nothing decisive about whether it is synthetic or authentic
6. A valid provenance record traces an image file. What still needs checking?
   • Whether the visual claim is true in context
   • Nothing; credentials prove every caption
   • Only whether the file is large
   • Only whether a detector also agrees
7. Why avoid a public accusation from one unvalidated detector flag?
   • Every flag is legally binding
   • The detector cannot be wrong on genuine images
   • False positives can harm real people, especially at low prevalence
   • The image becomes synthetic when flagged
8. Which evidence should accompany a consequential detection review?
   • Only the displayed confidence number
   • Original files, source history, tool version and independent corroboration
   • Only a cropped thumbnail
   • Only the detector’s marketing description