Design a small object segmentation and detection model for satellite imagery
Design an ML system capable of detecting and segmenting small objects (e.g., cars, roads, trees) in satellite images, addressing challenges such as class imbalance, low foreground-to-background ratio, and appropriate evaluation metrics for small objects.
Asked at:
KLA
Question Timeline
See when this question was last asked and where, including any notes left by other candidates.
Late July, 2026
KLA
Senior
The interviewer gives a satellite image showing roads, cars, trees etc. The problem is first how to design a segmentation/detection model. Then discuss strategies for segmenting small objects with very little foreground. Potential solutions are using higher resolution inputs, using focal loss, oversample small object image patches. Also consider which metrics to evaluate for small objects. One potential solution is to identify centroids of the object instead of evaluating mIOU.
Your account is free and you can post anonymously if you choose.