How do you handle varying lighting conditions and partial occlusion in object detection systems? What techniques at both data and model levels can improve robustness?

Sample answer preview

Varying lighting and occlusion represent two fundamental challenges in computer vision. Lighting affects the appearance of features the model learned during training. Occlusion removes parts of objects entirely, forcing the model to reason from incomplete information.