引導式測驗 · 中等 等級

使用 DeepLabCut 測驗進行動物姿勢估計

測試無標記標籤、使用者定義的關鍵點、可能性限制、遮蔽、會話測試、多動物身分、3D 視圖和行為聲明。

問題 1 的 8

What does “markerless” mean in DeepLabCut's animal-pose workflow?

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 “markerless” mean in DeepLabCut's animal-pose workflow?
    • Physical markers need not be attached, though training frames still need labels
    • The model needs no labeled examples of any body part
    • Every predicted coordinate is a direct physical measurement
    • The camera can recover hidden 3D joints from one view without assumptions
  2. Which landmarks can a researcher ask the system to track?
    • Only a fixed human skeleton designed by the original authors
    • Only objects with printed tracking tags
    • Only the center of the animal's bounding box
    • Body parts defined and labeled for the particular project
  3. An occluded paw has a low-likelihood predicted coordinate. How should a team use that output?
    • Treat the position as observed because a number was returned
    • Inspect the frame and uncertainty before using the coordinate in a movement measure
    • Interpret low likelihood as proof that the animal has no paw
    • Replace all other frames with the same coordinate
  4. Why is a test set of frames neighboring the training frames in one clip a weak generalization check?
    • Nearby frames always come from different animals
    • Video time order makes labels unnecessary
    • Near-identical appearance can leak across the split without testing new sessions
    • The test automatically covers every camera angle
  5. Two animals cross paths in a recording. What extra problem arises beyond locating paws and noses?
    • Every keypoint must become three-dimensional
    • Proving the animals have the same behavior
    • Choosing a different file format for each frame
    • Assigning each predicted part and trajectory to the correct individual
  6. What can changed lighting or a new cage background do to a trained pose model?
    • Increase errors because the new images differ from its labeled examples
    • Guarantee identical part accuracy after transfer
    • Remove the need to inspect any low-likelihood frames
    • Turn a 2D keypoint into a direct behavioral diagnosis
  7. For 3D pose, what do synchronized calibrated camera views add?
    • A guarantee that every hidden part is visible
    • A legal identity label for each animal
    • Multiple 2D observations whose geometry helps estimate depth
    • An automatic explanation of the animal's intent
  8. What does a nose trajectory alone fail to establish?
    • That coordinates were estimated across frames
    • The animal's inner state or cause of the movement
    • That a movement measure can be calculated from points
    • That the camera captured a visible nose in some frames