Q1
Walk me through how you would design a real-time object detection pipeline using a pre-trained model like YOLOv8 or Faster R-CNN. What considerations would you make for inference speed, memory usage, and accuracy trade-offs in a production environment?
Why they ask this:* They want to assess your understanding of modern detection architectures, optimization techniques, and practical deployment constraints that directly impact product performance.
Q2
Explain the differences between semantic segmentation and instance segmentation. When would you choose one approach over the other, and how would you evaluate which model architecture (U-Net, Mask R-CNN, DeepLab) is best suited for a specific business problem?
Why they ask this:* This tests your ability to distinguish between fundamental CV tasks, make architecture trade-off decisions based on requirements, and demonstrate knowledge of industry-standard frameworks.
Q3
Describe your experience with data annotation, labeling, and quality control processes. How do you handle imbalanced datasets, and what techniques (augmentation, class weighting, oversampling) have you used to improve model robustness?
Why they ask this:* They're evaluating whether you understand the data pipeline—often the biggest bottleneck in CV projects—and can manage the practical challenges of preparing training data at scale.
Q4
Walk through a computer vision project where you had to debug model performance issues. How did you identify whether the problem was in data preprocessing, model architecture, training hyperparameters, or post-processing?