Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Câu hỏi định dạng tập dữ liệu và chú thích COCO

Kiểm tra các liên kết danh mục hình ảnh COCO, hộp chiều cao x-y, phân đoạn phiên bản, chuyển đổi tọa độ, QA trực quan và phạm vi tập dữ liệu.

Đường dẫn hướng dẫn liên quanĐịnh dạng tập dữ liệu và chú thích Coco
Câu hỏi 1 của 8

A detection annotation has an image_id. What should that value link to?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. A detection annotation has an image_id. What should that value link to?
    • The category definition rather than an image
    • The model training run identifier
    • The image record containing the annotated object
    • The annotation’s own unique ID
  2. How is a standard COCO-style detection box commonly expressed?
    • Two opposing corners [x1,y1,x2,y2]
    • Upper-left x and y, then width and height
    • Center coordinates and half-width/height
    • A category ID followed by pixel count
  3. A source box is [x1,y1,x2,y2]. How should its width be converted?
    • Compute x2 minus x1
    • Use x2 directly as the width
    • Use x1+x2 as the width
    • Divide the image width by x2
  4. Why can one image have several annotation records?
    • One image must have a new ID per object
    • A category can occur only once per image
    • Each record represents an entire image rather than an instance
    • Each object instance can have its own category and geometry
  5. A warehouse dataset is converted to COCO JSON. What does that conversion itself establish?
    • Its records may be compatible with tools expecting the chosen COCO schema
    • Its image distribution now matches original COCO
    • Its category labels have been independently verified
    • Models trained on it will generalize to every warehouse
  6. A box stored in normalized coordinates is read as image pixels. What likely happens?
    • The box becomes larger because values are normalized
    • The category ID changes from integer to decimal
    • The parser automatically infers the missing pixel scale
    • The rendered box becomes tiny or misplaced
  7. Which check can reveal a category mapping error after conversion?
    • Verify only that the JSON parses
    • Draw sampled annotations and compare IDs, labels and image content
    • Inspect only total file size
    • Assume contiguous IDs imply correct category meaning
  8. What extra structure can instance segmentation supply beyond a box?
    • Guaranteed correctness of the category label
    • A verified camera-pose transform
    • A polygon or encoded mask describing an object’s pixels
    • A single image-level label replacing instances