Guide-linked quiz · Hard Level
6D Object Pose Estimation Quiz
Test 6D translation and rotation, camera-relative frames, model correspondences, symmetry-aware scoring, and pose use in robots and AR.
Question 1 of 8
A detector gives a tight 2D box around a tool. Which information is still needed for a gripper to approach it?
Preview every question8 questions
Answers and explanations appear as you take the quiz.
- A detector gives a tight 2D box around a tool. Which information is still needed for a gripper to approach it?
- Only the class label and detector confidence
- Only a 2D mask around the visible pixels
- Only distance to the nearest surface without orientation
- The tool’s 3D position and orientation in a known frame
- In a rigid 6D pose, what do the six degrees describe?
- Three translation and three rotational degrees
- Three translations and three independent scale factors
- Six articulation angles for a deformable object
- Three colors plus three depth channels
- Why should camera intrinsics be known when projecting a 3D model onto an image?
- They supply the object’s rigid transform relative to the camera
- They directly remove occlusion from the image
- They map camera-frame points to image coordinates
- They determine the object’s real-world identity
- An unmarked cylinder looks the same after rotation around its axis. How should evaluation handle this?
- Penalize every angle other than one stored annotation
- Treat visually or geometrically equivalent rotations as equivalent where the task allows
- Ignore all translation errors for symmetric objects
- Assume a unique orientation from geometry that is unchanged
- What can RGB-D data add to an RGB-only pose estimate?
- Measured range cues that may help constrain object geometry and translation
- An exact pose on every reflective surface
- A built-in 3D model for every unknown object
- A semantic identity label for each pixel
- A pose method trained on synthetic renders is tested on real camera images. Which risk deserves measurement?
- Real camera appearance is guaranteed to match rendering
- Synthetic training removes the need for calibration
- A domain gap in appearance and sensor characteristics
- Rigid pose has extra degrees only in real scenes
- Why might box-overlap accuracy be insufficient for robotic pose evaluation?
- Good 2D overlap guarantees correct 3D rotation
- Two boxes can overlap while orientations or 3D translations differ enough to spoil a grasp
- The box alone specifies a collision-free grasp
- Pose error matters only to image compression
- What assumption limits a known-model 6D pose method on an unfamiliar object?
- A known model can be recovered from the class name alone
- RGB input cannot be used with known models
- Object coordinates cease to exist without depth
- It may lack the correct 3D shape and scale needed for matching
Keep testing yourself
More quizzes picked for your level.