Computer Vision Interview 20 essential Q&A Updated 2026
YOLO

YOLO: 20 Essential Q&A

You Only Look Once—grids, anchors, and the push for real-time detection.

~12 min read 20 questions Advanced
one-stagegridanchorslatency
1 What does YOLO mean? 📊 medium
Answer: You Only Look Once: single forward pass predicts boxes and classes—treats detection as regression from a grid of cells.
2 YOLOv1 grid idea? 📊 medium
Answer: Image split into S×S cells; cell responsible for object whose center falls in it—predicts B boxes + class distribution per cell.
3 YOLOv1 loss components? 🔥 hard
Answer: Coordinate regression (with sqrt w,h trick), confidence (IoU weighted), classification CE—λ weights balance localization vs no-object cells.
4 When did anchors appear? 📊 medium
Answer: YOLOv2+ uses k-means anchor priors on dataset boxes—predict offsets instead of raw sizes for stability.
5 IoU in training? 📊 medium
Answer: Assign anchors/cells to GT by best IoU; some versions ignore preds below IoU threshold for classification to reduce conflict.
6 Post-processing? ⚡ easy
Answer: Like other detectors: NMS on decoded boxes with class-wise scores—some variants use DIoU-NMS or soft-NMS.
7 Objectness vs class? ⚡ easy
Answer: Objectness = is there an object in this anchor; class = which class—decoupled in many heads (obj * class prob = final score).
8 Multi-scale YOLO? 📊 medium
Answer: Later versions predict at multiple feature map scales (e.g. large/small stride) to catch objects of different sizes—similar spirit to FPN.
9 Path aggregation? 📊 medium
Answer: Models like YOLOv4 use PANet-style bottom-up path after top-down FPN for richer multi-scale features.
10 YOLOv5/v8 / Ultralytics? ⚡ easy
Answer: Popular PyTorch implementations with training zoo, export, and deployment tooling—interview “practical YOLO” often means this ecosystem.
11 Deploy on edge? 📊 medium
Answer: Export to ONNX, TensorRT, CoreML—quantize INT8 for speed; validate mAP drop after conversion.
12 Small objects? 📊 medium
Answer: Higher-res input, smaller stride heads, copy-paste aug, or tiling—same fundamentals as other detectors.
13 Crowded objects? 🔥 hard
Answer: Grid responsibility and NMS can struggle—improved assignment (e.g. ATSS-style ideas in some detectors) and better NMS help.
14 Common augmentations? 📊 medium
Answer: Mosaic, mixup, HSV jitter, random scale—strong aug standard in modern YOLO training recipes.
15 mAP vs FPS tradeoff? ⚡ easy
Answer: Larger model and image size ↑ mAP, ↓ FPS—choose for product SLA (latency vs accuracy).
16 YOLO vs SSD? 📊 medium
Answer: Both one-stage; SSD uses multi-scale default boxes on VGG features; YOLO family evolved different heads and assignment—both real-time capable.
17 YOLO vs RetinaNet? 📊 medium
Answer: RetinaNet introduced focal loss for dense classification imbalance; YOLO uses different obj loss weighting—both dense predictors.
18 Tiling satellite / huge images? 📊 medium
Answer: Split image, run YOLO per tile with overlap, merge + NMS—handle boundary duplicates.
19 Rotated boxes? 🔥 hard
Answer: Variants predict angle θ or use rotated IoU—needed for aerial/text detection.
20 Real-time on CPU? 📊 medium
Answer: Choose nano/tiny backbones, reduce input size, INT8—expect large accuracy gap vs GPU server models.

YOLO Cheat Sheet

Idea
  • Single forward
  • Dense preds
Train
  • Anchors (v2+)
  • Multi-scale heads
Ship
  • NMS
  • TensorRT / ONNX

💡 Pro tip: One-stage = dense predictions + clever assignment + NMS.

Full tutorial track

Go deeper with the matching tutorial chapter and code examples.