Image-to-3D Model Generation Quiz

Answers and explanations appear as you take the quiz.

1. What does feed-forward reconstruction mean in image-to-3D?
   • The model produces the 3D result in one pass through a trained network, in seconds
   • The model optimizes each object for hours
   • A human sculpts the object from the photo
   • The model scans the object with a laser
2. What representation does LRM's transformer map image features into?
   • A MIDI file
   • A point cloud from a depth sensor
   • A triplane of three axis-aligned feature planes
   • A single 2D depth map only
3. Which organizations released TripoSR in March 2024?
   • Adobe and Google
   • Stability AI and Tripo AI
   • Meta and Sony
   • OpenAI and Microsoft
4. How is a mesh typically extracted from the predicted 3D field?
   • By hand tracing
   • By photogrammetry
   • By UV unwrapping alone
   • With the marching cubes algorithm
5. What did Zero-1-to-3 contribute to image-to-3D methods?
   • Laser scanning of objects
   • Automatic retopology
   • Using a diffusion model to imagine new views of an object
   • A licensing framework for 3D assets
6. Why is the back of a reconstructed object a guess?
   • A single photo does not contain the hidden side, so the model infers it from training patterns
   • The model deliberately randomizes it
   • Meshes cannot have backs
   • The GPU runs out of memory
7. Why are dimensions from a single-image 3D model not trustworthy?
   • Meshes are always saved in inches
   • The model rounds to whole numbers
   • Marching cubes shrinks objects
   • Scale cannot be determined from one image
8. Which preprocessing step typically improves single-image 3D results?
   • Adding a busy background
   • Removing the background and centering the object
   • Converting the photo to black and white
   • Cropping off part of the object