DepthAnything Monocular Depth Quiz

Answers and explanations appear as you take the quiz.

1. What does 'monocular' depth estimation mean?
   • Predicting depth from a single image
   • Using two cameras for stereo vision
   • Using a LiDAR sensor
   • Measuring depth with sound waves
2. What was DepthAnything's main strategy for achieving strong generalization?
   • Hand-labeling a small high-quality dataset
   • Training on roughly 62 million auto-labeled unlabeled images
   • Using only synthetic game engine data
   • Relying on stereo camera pairs
3. What backbone encoder does DepthAnything build on?
   • A ResNet-18 CNN
   • A recurrent LSTM network
   • The DINOv2 vision transformer
   • A YOLO detector
4. What kind of depth does the original DepthAnything primarily output?
   • Exact distance in meters
   • Only binary foreground/background
   • GPS coordinates
   • Relative depth (nearer vs. farther)
5. How did DepthAnything V2 improve fine detail and edges?
   • By training the teacher on synthetic data with perfect ground truth, then distilling
   • By adding more real noisy labels
   • By removing the transformer encoder
   • By using lower-resolution inputs
6. Which application is NOT a typical use of monocular depth maps?
   • Portrait-mode background blur
   • Encrypting network traffic
   • ControlNet depth conditioning
   • 2D-to-3D photo conversion
7. What dense prediction head does the Depth Anything guide pair with its DINOv2 encoder?
   • A Marching Tetrahedra mesh extractor
   • A CLIP text-image similarity head
   • A DPT-style dense prediction head
   • A causal language-model decoder
8. How does Depth Anything V2 use a teacher model with unlabeled real photos?
   • The teacher converts each photo into a 3D mesh
   • The teacher supplies only image captions to a language model
   • The teacher replaces all synthetic depth labels with manually measured distances
   • The teacher generates pseudo-depth labels that are used to train student models