Animacja portretowa z quizu dotyczącego pojedynczego zdjęcia

Odpowiedzi i wyjaśnienia pojawiają się w miarę rozwiązywania quizu.

1. W jaki sposób First Order Motion Model uzyskuje swoje punkty kluczowe?
   • From manually labeled facial landmarks
   • From a 3D scan of the person
   • It learns them without labels during training
   • From the audio track
2. Jaki jest cel mapy okluzji?
   • To mark regions the source image cannot supply, so they must be painted in
   • To blur the background on purpose
   • To detect the speaker's language
   • To choose the output frame rate
3. Jaka jest korzyść z relatywnego przenoszenia ruchu?
   • It doubles the frame rate
   • The source keeps its own face shape instead of taking on the driver's
   • It removes the need for a driving video
   • It makes the head turn farther
4. Co SadTalker przewiduje na podstawie dźwięku?
   • Raw pixel values of the mouth
   • A new background
   • Transkrypcja tekstowa
   • 3D Morphable Model coefficients for expression and head pose
5. What does LivePortrait's stitching module do?
   • It pastes the animated face back into the full image without misaligned shoulders
   • It merges two different people's faces
   • It stitches audio clips together
   • It converts video to GIF
6. Why do large head turns usually fail in single-photo animation?
   • The GPU runs out of memory
   • Head turns are disabled by default
   • The photo has no information about unseen regions like the side of the head
   • Audio cannot describe rotation
7. Why do keypoint methods warp feature maps rather than raw pixels?
   • Pixels cannot be moved
   • It lets the decoder repair stretching and fill unseen regions
   • Feature maps are always higher resolution
   • It removes the need for keypoints
8. In implicit 3D keypoint methods, what does the driving frame contribute?
   • The canonical keypoints encoding identity
   • The background texture
   • The skin color
   • Head rotation, translation and expression deformations