Saral Shiksha Yojna
Courses/Computer Vision

Computer Vision

CSE471
Prof. Makarand Tapaswi + Prof. Charu Sharma•Spring 2025-26•4 credits
Sample Papers/Mock Paper 7 — Rapid Drill (50 short questions across all topics, 90 min)

Mock Paper 7 — Rapid Drill (50 short questions across all topics, 90 min)

Duration: 90 min • Max marks: 100

All Questions (2 marks each, 50 questions, 100 marks)

100 marks
  1. 1.Modal vs amodal bounding box detection.2 m
  2. 2.256×256 image, downsampled by stride-2 convolutions four times. Resulting spatial dimension?2 m
  3. 3.Vanishing gradient problem and which architecture directly addressed it for deep CNNs?2 m
  4. 4.In Sobel edge detection, what does G_x detect — horizontal or vertical edges?2 m
  5. 5.Convolution Theorem.2 m
  6. 6.What does Otsu's method optimise?2 m
  7. 7.What does cv2.dilate do mathematically in one sentence?2 m
  8. 8.Formula for mean Average Precision (mAP).2 m
  9. 9.In Faster R-CNN, what does the RPN output and what is its loss function?2 m
  10. 10.Why does YOLO use √w and √h in its box-size loss?2 m
  11. 11.Difference between Dice and IoU.2 m
  12. 12.Focal Loss formula and what it solves.2 m
  13. 13.U-Net skip connections — additive or concatenative?2 m
  14. 14.What does RoI Align fix compared to RoI Pool?2 m
  15. 15.Transposed convolution — also known as what, and is this name accurate?2 m
  16. 16.What is a Part Affinity Field in OpenPose?2 m
  17. 17.Difference between PCK and PCKh.2 m
  18. 18.SMPL parameters: which controls shape, which controls pose, dimensions?2 m
  19. 19.Scaled dot-product attention formula.2 m
  20. 20.Why must Transformers use positional encoding?2 m
  21. 21.In ViT, where does the [CLS] token attend, and what is its role?2 m
  22. 22.Patch count for a 384×384 image with patch size 14×14.2 m
  23. 23.Swin Transformer's key innovation over vanilla ViT.2 m
  24. 24.What does RoPE stand for and what does it do?2 m
  25. 25.CLIP with N pairs per batch: how many positive vs negative pairs?2 m
  26. 26.What is SigLIP different from CLIP?2 m
  27. 27.DINO teacher EMA update rule.2 m
  28. 28.DINO's two anti-collapse mechanisms.2 m
  29. 29.What does MAE stand for and what is its mask ratio?2 m
  30. 30.What does JEPA predict, and how is this fundamentally different from MAE?2 m
  31. 31.3DGS parameters per Gaussian with SH degree 3.2 m
  32. 32.Why is 3DGS's Σ parameterised as R·S·Sᵀ·Rᵀ?2 m
  33. 33.Three pillars of 3D Gaussian Splatting.2 m
  34. 34.PaliGemma's connector type and dimensions.2 m
  35. 35.Qwen2-VL's M-RoPE — what does M stand for and what does it encode?2 m
  36. 36.Typical kernel shape in a 3D CNN (e.g., I3D).2 m
  37. 37.I3D's inflation trick.2 m
  38. 38.Describe SlowFast's two pathways.2 m
  39. 39.What does TimeSFormer do with attention?2 m
  40. 40.Why is permutation invariance important for point-cloud networks like PointNet?2 m
  41. 41.Which operation in PointNet provides permutation invariance?2 m
  42. 42.What are PointNet's 'critical points'?2 m
  43. 43.DGCNN's EdgeConv — key input feature?2 m
  44. 44.What makes DGCNN's graph 'dynamic'?2 m
  45. 45.MeshCNN operates on which mesh element, and why?2 m
  46. 46.Difference between classification and regression in terms of loss functions.2 m
  47. 47.Model has 90% accuracy on a binary task where 90% are negative. Is this good?2 m
  48. 48.In ROC analysis, what do TPR and FPR mean?2 m
  49. 49.Define precision and recall.2 m
  50. 50.F1 score formula.2 m

Track your attempt locally — score and time are recorded in your browser. (Coming soon: timed-attempt mode.)