引導式測驗 · 中等 等級

多邊形與遮罩分割標註測驗

比較多邊形和像素遮罩、全景標籤、COCO 編碼、重疊檢查、提示 SAM 說明和分段限制。

問題 1 的 8

為什麼註釋團隊會選擇使用蒙版而不是多邊形來標記照片中人物的頭髮?

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. 為什麼註釋團隊會選擇使用蒙版而不是多邊形來標記照片中人物的頭髮?
    • Hair has irregular fine-grained edges that straight polygon lines would blur
    • Masks are always faster to draw than polygons
    • Polygons cannot be stored in JSON format
    • Masks require less storage than polygons in every case
  2. 2023 年發布的原始 SAM 為分段工作流程提供了什麼?
    • A guarantee that all masks are correct
    • A promptable model that proposes masks from inputs such as points or boxes
    • An annotation format replacing COCO
    • A universal text-only segmentation endpoint
  3. 全景分割與僅標示汽車和行人等前景物體有何不同?
    • It only labels moving objects
    • It only works on grayscale images
    • It assigns a label to every pixel in the image, including background regions like sky and road
    • It removes the need for polygon annotation entirely
  4. 在COCO註解格式中,多邊形頂點通常如何表示?
    • As a single average pixel color
    • As a run-length encoded bitmap only
    • As a text description of the shape
    • As an ordered list of (x, y) coordinates
  5. 根據指南,遮罩註解中的遊程編碼(RLE)有何用途?
    • Compressing long runs of identical pixel values in a binary mask
    • Recording the order in which labelers worked
    • Encoding polygon vertex coordinates
    • Measuring annotator agreement
  6. How is Intersection over Union (IoU) commonly used in segmentation quality control?
    • To measure typing speed of annotators
    • To compare a submitted mask's overlap against a reference mask and flag low agreement
    • To convert polygons into bounding boxes
    • To decide which image format to use for storage
  7. Why does video segmentation remain harder to automate than single-image segmentation?
    • Video files cannot be opened by segmentation models
    • Videos never contain more than one object
    • Masks must stay consistent across frames as objects move, which single-image tools do not handle well
    • Polygon annotation is impossible on video
  8. For which type of object would a polygon annotation likely match mask accuracy while using less storage?
    • Smoke drifting in the wind
    • A person's individual eyelashes
    • Foliage on a tree
    • A parked car with mostly straight and smooth edges