引導式測驗 · 中等 等級

使用 SMPL 測驗進行 3D 人體網格恢復

測試 SMPL 形狀和姿勢參數、單影像 HMR、重投影模糊度、隱藏表面、服裝限制、3D 驗證和隱私。

相關引導路徑3D 人體網格恢復範例
問題 1 的 8

What does the basic SMPL model provide to a mesh-recovery system?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What does the basic SMPL model provide to a mesh-recovery system?
    • A single fixed body mesh whose pose and proportions cannot vary
    • An exact clothed-surface mesh including every garment fold
    • A measured 3D scan of each input person's full body
    • A parametric articulated body mesh with a consistent structure
  2. In the guide's SMPL explanation, what do pose parameters control?
    • Overall body proportions represented by shape coefficients
    • The articulation of body joints and resulting mesh configuration
    • Camera position and projection relative to the body
    • A fixed vertex order independent of the body's motion
  3. What kind of input did the cited Human Mesh Recovery research use for its single-image estimate?
    • A complete 3D body scan required for every prediction
    • Synchronized multi-view footage required for every estimate
    • One RGB image with learned body and camera inference
    • A separate depth map from a range camera for every frame
  4. Why does a mesh fitting the visible 2D silhouette not prove exact 3D body dimensions?
    • Different depth, pose and camera configurations can project similarly
    • A silhouette always exposes the back of the body
    • A closer camera and smaller body necessarily make different 2D images
    • SMPL stores every hidden depth measurement in the image
  5. Which visible details are outside the basic SMPL body mesh's promised geometry?
    • Body pose and learned body shape
    • Loose garments and hair strands
    • The model's skeletal articulation
    • The mesh's shared vertex structure
  6. A person's arm is hidden behind the torso in one photograph. What should a reviewer infer about the recovered arm pose?
    • Occlusion makes any possible arm pose equally likely
    • It was observed from the camera's hidden-side view
    • The mesh proves the arm was in exactly one physical position
    • It is an uncertain model estimate because the arm is not directly visible
  7. Which test is stronger than checking only a plausible 2D overlay?
    • Comparing recovered 3D pose and shape with independent reference data where available
    • Reusing the same person's near-identical frames as independent test cases
    • Comparing only the displayed mesh color with the photograph
    • A viewer's impression that the rendering looks smooth
  8. Why do consent and retention controls apply to mesh-recovery outputs as well as photographs?
    • A mesh is always publicly anonymous because it has no color
    • Derived parameters can never be linked to the input person
    • Images and inferred body measurements can reveal personal information
    • A prediction automatically authorizes future reuse of the image