引导式测验 · 中等 等级

人体姿势估计测验

查看 2D 姿势关键点、自上而下和自下而上的管道、零件亲和力场以及图像平面估计无法确定的内容。

相关引导路径人体姿势估计
问题 1 的 8

人体姿态估计主要检测什么？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. 人体姿态估计主要检测什么？
    • 身体关节或关键点的位置
    • 衣服的颜色
    • 此人的姓名
    • 图像文件大小
  2. 自上而下的姿态估计方法如何工作？
    • 他们立即检测所有关节，然后将它们分组
    • 他们首先检测每个人，然后找到盒子内的关节
    • 他们从不使用边界框
    • 它们仅适用于单个静态关节
  3. OpenPose 预测什么来估计身体部位位置？
    • A direct 3D skeleton with depth
    • A single bounding box per frame
    • Confidence maps for candidate body-part locations
    • A sequence of spoken joint names
  4. OpenPose 解析器使用什么步骤将候选身体部件关联成完整姿态？
    • A face-recognition classifier
    • A top-down detector for every limb
    • A language-model prompt
    • A set of bipartite matchings scored with Part Affinity Fields
  5. OpenPose 论文讨论了顶部向下姿态估计在拥挤场景中的哪种失败模式？
    • A person detector may miss people close together, and the pose stage cannot recover them
    • It always detects more people than are present
    • It cannot estimate any 2D keypoints
    • It requires a 3D depth camera
  6. Which additional keypoint groups did the OpenPose paper include in its system?
    • Only facial landmarks
    • Body, foot, hand and facial keypoints
    • A 3D skeleton with medical diagnoses
    • Only fingertip positions
  7. What information do Part Affinity Fields encode to connect keypoints?
    • A person’s identity from face recognition
    • Depth in meters for every body joint
    • Vectors indicating limb position and orientation across the image
    • The camera exposure settings
  8. What can a 2D pose model conclude directly from image keypoints?
    • A person’s medical diagnosis
    • Exact real-world joint depth without additional information
    • The person’s intentions
    • Approximate joint locations in image coordinates