1What was the first 3D human face rendering method, pioneered by Frederic Ira Parke in 1972?
Blend Shapes
2What type of technique is Blend Shapes?
An interpolation technique (blending between target expression meshes).
3What is the main drawback of Blend Shapes?
A weakness in expressing subtle variety.
4Who developed the Facial Action Coding System (FACS)?
Paul Ekman
5What does FACS anatomically split facial expressions into?
Specific muscle movements called Action Units (AUs).
6According to FACS, which AUs correspond to 'Happiness'?
AU 6 + AU 12
7According to FACS, which AUs correspond to 'Sadness'?
AU 1 + AU 4 + AU 15
8What is a key challenge when animating from speech, related to the influence of nearby sounds?
Coarticulation
9What two inputs does the JALI (Jaw and Lips) model use?
Audio and a text transcript.
10What algorithm does JALI use for "forced alignment" (syncing phonemes to audio)?
HMM (Hidden Markov Model)
11What NVIDIA model generates 3D facial animation in real-time from only an audio stream?
Audio2Face (A2F)
12What deep learning architecture does Audio2Face (A2F) use?
A wav2vec2.0 architecture.
13What is a key limitation of Audio2Face (A2F)?
It can be inaccurate with background noise and cannot generate natural 'idle' or listening motions.
14What is the VOCA model good at capturing?
Synthesizing different 3D speaking styles from audio.
15What model architecture does FaceFormer use for speech-driven animation?
Transformers
16What specific challenge does the "Learning to Listen" model address?
Animating the non-deterministic facial reactions of a listener in a conversation.
17What novel rendering technique does the audio-driven UniGAHA model use?
Gaussian Splatting
18What is the ProbTalk3D model (from Utrecht University) good at generating?
Non-deterministic and emotion-controllable 3D facial animation.
19What does 3DMM stand for?
3D Morphable Models
20What are 3DMMs (like FLAME) good at disentangling?
The core components of a face (e.g., shape, expression, pose).
21What is the RingNet model good at?
Regressing a 3D face shape and expression from a single 2D image.
22What is the DECA model good at learning from?
"In-the-wild" (uncontrolled) images.
23What does the SAMPL-X model capture from a single image?
The full expressive body (3D hands, face, and body).
24What was the massive facial animation challenge for the game Cyberpunk 2077?
Localizing facial animation for 10 different voice-over languages.
25What model did Cyberpunk 2077 use to solve its animation localization problem?
JALI
26What did Utrecht University's 'Realism' study find?
Higher appearance and animation realism led to higher perceived social presence and attractiveness.
27What 'uncanny valley' effect did the 'Familiarity' study on 'digital doubles' find?
Participants could identify their own digital doubles but consistently disliked them, especially realistic versions.
28In the 'Familiarity' study, who were people more forgiving of?
The digital doubles of acquaintances or unknown persons.