引導式測驗 · 中等 等級

電腦視覺工程師職業測驗

測試電腦視覺任務定義、資料集和度量推理、遷移學習、特定於角色的資格以及部署權衡。

問題 1 的 8

哪種輸出是物件偵測特有,而非影像分類?

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. 哪種輸出是物件偵測特有,而非影像分類?
    • Classification requires bounding boxes for every object
    • Object locations or bounding boxes paired with class labels
    • One or more labels for the whole image
    • A pixel-level class mask for each object
  2. 在驗證分數意外地高時,工程師應該檢查什麼?
    • Whether the model has enough deployment replicas
    • Whether the color profile matches the dashboard
    • Whether the optimizer uses a larger learning rate
    • Possible leakage or overlap between data splits
  3. 對於視覺分類器,什麼時候召回率(recall)比精確率(precision)更重要?
    • When missing a positive case is costlier than a false alarm
    • When false alarms are the only harmful error
    • When the dataset has no positive labels
    • When all thresholds produce the same predictions
  4. 引用的 PyTorch Mask R-CNN 教程展示了什麼?
    • Replacing all image labels with embeddings
    • Deploying an autonomous vehicle system
    • Fine-tuning a pretrained model for detection and instance segmentation
    • Training a universal vision model from scratch
  5. 當目標裝置上的推論速度過慢時,電腦視覺工程師可以怎麼做?
    • Change the evaluation labels until runtime improves
    • Choose a different metric without measuring latency
    • Assume cloud inference is always acceptable
    • Profile the pipeline and evaluate model or deployment optimizations
  6. What qualification does the cited Google posting list for its specific role?
    • Programming experience in Python or C++
    • A required doctorate for every vision engineer
    • Experience only with JavaScript
    • A guarantee of prior autonomous-vehicle work
  7. How does image segmentation differ from object detection in the guide?
    • Detection assigns a label to every pixel
    • They are identical outputs with different names
    • Segmentation assigns labels at pixel or instance level
    • Segmentation returns only one image-level class
  8. How should candidates treat the guide’s practice prompts?
    • As replacements for reading job requirements
    • As preparation examples, not a promise about an employer’s interview
    • As leaked questions that all companies use
    • As a required coding test for every role