Vision Tasks for Deep Learning Engineers
Vision tasks are the standard problems solved with image models: classification and representation learning, object detection that draws boxes, segmentation that labels pixels, and pose estimation, tracking and re-identification across video. Each has its own output format, loss and metrics, such as accuracy, mAP and IoU.
Stage: Computer Vision · Part of the Deep Learning Engineer roadmap · Last reviewed 19 November 2026
What to learn in Vision Tasks
- Compute intersection over union and use it to match boxes
- Explain anchor-based and anchor-free detectors
- Distinguish semantic, instance and panoptic segmentation
- Evaluate a detector with mean average precision
- Use self-supervised pretraining to learn image representations
- Link detections across frames to track objects
Vision Tasks interview questions
What is mean average precision in object detection?
For each class, average precision is the area under the precision-recall curve, where a detection counts as correct if its IoU with a ground truth box passes a threshold. mAP averages this over classes, and sometimes over several IoU thresholds.
What is non-maximum suppression?
Non-maximum suppression removes duplicate detections. It sorts boxes by confidence, keeps the best, and discards others that overlap it above an IoU threshold, repeating for the remaining boxes. Some modern detectors avoid it by design.
Where Vision Tasks fits in the Deep Learning Engineer roadmap
Vision Tasks belongs to the Computer Vision stage (advanced level), which takes about 4 hours. See and understand.
Other units in this stage
Back to the full Deep Learning Engineer roadmap