Magic3D Text-to-3D Pipeline Quiz

Answers and explanations appear as you take the quiz.

1. What overall structure defines the Magic3D pipeline?
   • A two-stage coarse-to-fine process
   • A single end-to-end diffusion pass
   • Pure GAN training
   • A retrieval lookup from a 3D database
2. Which representation does the fine stage optimize for crisp detail?
   • A dense voxel grid
   • A textured triangle mesh via differentiable rasterization
   • A raw point cloud
   • A 2D image only
3. What speeds up the coarse stage's geometry estimation?
   • Running on CPU only
   • Skipping the diffusion prior
   • A hash-grid neural field (Instant-NGP style)
   • Using full 1024-resolution rendering
4. What runtime comparison did the Magic3D paper report against DreamFusion?
   • About 50× slower
   • No runtime difference in any tested setup
   • About 1,000× faster in every case
   • About 2× faster in the reported experiments
5. Why does the fine stage run its diffusion prior in latent space?
   • To supervise high-resolution detail efficiently and cheaply
   • To remove the need for prompts
   • To convert the model to audio
   • To avoid using any GPU
6. What advantage does Magic3D's mesh output give downstream users?
   • It can only be viewed in the browser
   • It is directly editable in standard 3D graphics tools
   • It cannot be textured
   • It must be re-scanned before use
7. What user-study result did the Magic3D paper report when comparing its outputs with DreamFusion?
   • Every rater preferred DreamFusion
   • Exactly half of all users preferred each method in every task
   • 61.7% of raters preferred Magic3D in the reported comparison
   • No human comparison was conducted
8. What additional input modality can Magic3D use to condition 3D generation?
   • A telephone codec score
   • A dependency parse tree
   • A robot torque sequence
   • A reference image