Q1
Walk me through how you would design and implement a real-time object detection pipeline using a pre-trained model like YOLOv8 or Faster R-CNN. What considerations would you make for latency, accuracy, and hardware constraints?
Why they ask this:* They want to assess your understanding of modern detection architectures, inference optimization, and practical deployment trade-offs that are critical for production computer vision systems.
Q2
Explain the difference between semantic segmentation and instance segmentation. When would you choose one over the other, and what are the computational implications of each approach?
Why they ask this:* This tests your foundational knowledge of core computer vision tasks, your ability to match algorithms to business problems, and your awareness of performance considerations.
Q3
You're working with a dataset where your model shows strong performance on test data but poor performance in production. What are the common causes of this domain shift problem, and what techniques would you use to diagnose and mitigate it?
Why they ask this:* They're evaluating your awareness of real-world challenges like data drift and distribution shift—critical issues that mid-level engineers must handle independently in production environments.
Q4
Describe your experience with data augmentation strategies for computer vision tasks. How do you decide which augmentations to apply, and have you used libraries like albumentations or imgaug? What are the risks of over-augmentation?