Saral Shiksha Yojna
Courses/Computer Vision

Computer Vision

CSE471
Prof. Makarand Tapaswi + Prof. Charu Sharma•Spring 2025-26•4 credits
Sample Papers/Original — Mid-Sem Practice (Units 1–6)

Original — Mid-Sem Practice (Units 1–6)

Duration: 90 min • Max marks: 50

Section A — Short Answers (each 3 marks)

15 marks
  1. 1.Define IoU and GIoU. State the range of GIoU and one situation where GIoU has a non-zero gradient but IoU does not.3 m
  2. 2.State the YOLO v1 output tensor shape and explain each dimension for PASCAL VOC.3 m
  3. 3.Difference between RoI Pool and RoI Align. Which is used in Mask R-CNN and why?3 m
  4. 4.Explain heatmap regression for pose estimation in one paragraph.3 m
  5. 5.Write the per-Gaussian parameter count in 3DGS as a sum.3 m

Section B — Long Answers (each 7 marks)

21 marks
  1. 1.Trace the R-CNN → Fast → Faster R-CNN evolution. For each step, identify the bottleneck removed and the mechanism replacing it. Include the role of the RPN and its anchor scheme.7 m
  2. 2.Explain Part Affinity Fields in OpenPose. Write the line-integral score for a candidate (A, B) pair and describe how bipartite matching assembles individuals.7 m
  3. 3.Derive (informally) why scaled dot-product attention divides by √dₖ. State the assumed input distribution and where softmax saturation would occur without scaling.7 m

Section C — Application / Algorithm trace (each 7 marks)

14 marks
  1. 1.Given 5 detections with confidence scores [0.9, 0.85, 0.7, 0.6, 0.4] and the following pairwise IoUs (D₁D₂ = 0.6, D₁D₃ = 0.2, D₁D₄ = 0.55, D₁D₅ = 0.1, D₂D₃ = 0.15, D₂D₄ = 0.25, D₂D₅ = 0.05, D₃D₄ = 0.3, D₃D₅ = 0.45, D₄D₅ = 0.1), perform NMS with τ = 0.5. Show the kept-and-suppressed table step-by-step.7 m
  2. 2.Compute the rough parameter count for ViT-B/16 (L=12, d_model=768, d_ff=3072, patch=16, image=224). Show per-layer breakdown and total.7 m

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