Q1
Walk me through how you would design a real-time object detection pipeline for an autonomous vehicle. What trade-offs would you consider between model accuracy, inference latency, and computational resources?
Why they ask this:* They want to assess your understanding of end-to-end system design, knowledge of optimization techniques, and ability to balance competing constraints in production computer vision systems.
Q2
Explain the differences between CNN architectures like ResNet, EfficientNet, and Vision Transformers. When would you choose one over the others for a new project, and what factors influence your decision?
Why they ask this:* They're testing your depth of knowledge in modern architectures, ability to evaluate trade-offs, and whether you stay current with emerging techniques in the field.
Q3
Describe your experience with data annotation, augmentation, and handling class imbalance in computer vision datasets. What techniques have you used to improve model robustness with limited labeled data?
Why they ask this:* They want to understand your practical experience with the data pipeline—a critical bottleneck in AI projects—and your problem-solving approach when resources are constrained.
Q4
How would you approach debugging a computer vision model that performs well on validation data but poorly in production? What tools and methodologies would you use?