Computer Vision Practice

CV Practice Exercises (Topic-Wise)

Each block gives a quick drill plus direct links to the tutorial chapter and 15-question MCQ for that topic—aligned with the sidebar curriculum. Use this page for spaced revision, interview prep, and exam-style checks.

How to practice

Spend 10–20 minutes per topic: answer the drill mentally or in a notebook, skim the tutorial for gaps, then run the MCQ once cold and once after review. Pair with hands-on projects for coding muscle memory.

Fundamentals & orientation

What is computer vision?

Drill: In one paragraph, contrast CV with image processing and list three real deployments.

Tutorial 15 MCQs

Hands-on project ideas

Drill: Pick one beginner build from the hub and list inputs, outputs, and one metric you would report.

Projects hub

Image processing basics

Drill: Given H×W×3 RGB, state memory size in bytes (uint8) and what changes when converting to grayscale.

Tutorial 15 MCQs

Color spaces

Drill: Explain why HSV is often used to segment a colored object under varying light.

Tutorial 15 MCQs

Image transformations

Drill: Compare affine vs perspective transforms; when do parallel lines stay parallel?

Tutorial 15 MCQs

Preprocessing & enhancement

Image filtering / convolution

Drill: Relate kernel size to blur strength and border handling choices.

Tutorial 15 MCQs

Edge detection

Drill: Order Sobel → gradient magnitude → Canny conceptually; what does hysteresis fix?

Tutorial 15 MCQs

Thresholding

Drill: When does Otsu fail? Name one adaptive alternative.

Tutorial 15 MCQs

Morphological operations

Drill: Describe opening vs closing on a binary text mask with salt noise.

Tutorial 15 MCQs

Histogram equalization

Drill: Why can global equalization over-enhance noise; what does CLAHE change?

Tutorial 15 MCQs

Feature detection & descriptors

Feature detection intro

Drill: Define keypoint vs descriptor; why match ratio tests?

Tutorial 15 MCQs

Harris corners

Drill: What do eigenvalues of the structure tensor imply about local intensity variation?

Tutorial 15 MCQs

SIFT

Drill: Outline scale-space extrema → orientation → descriptor at a high level.

Tutorial 15 MCQs

ORB

Drill: Why is ORB attractive for mobile compared to SIFT?

Tutorial 15 MCQs

HOG

Drill: How do cells, blocks, and normalization support pedestrian templates?

Tutorial 15 MCQs

Segmentation

Segmentation overview

Drill: Compare region growing vs graph cuts in one sentence each.

Tutorial 15 MCQs

Semantic segmentation

Drill: Why is per-pixel cross-entropy common; what does skip fusion fix?

Tutorial 15 MCQs

Instance segmentation

Drill: How do masks differ from semantic labels when two objects share a class?

Tutorial 15 MCQs

Object detection

Detection intro

Drill: Write IoU for two axis-aligned boxes and give a typical positive threshold.

Tutorial 15 MCQs

R-CNN family

Drill: Contrast two-stage proposals+refine vs single-shot latency tradeoff.

Tutorial 15 MCQs

YOLO

Drill: Explain grid-based prediction and why multiple anchors per cell help.

Tutorial 15 MCQs

RetinaNet / focal loss

Drill: What imbalance problem does focal loss address in dense detection?

Tutorial 15 MCQs

Tracking

Tracking basics

Drill: Differentiate detection in every frame vs short-term association.

Tutorial 15 MCQs

Kalman filter

Drill: State predict vs measurement update in one line each for a bounding box center.

Tutorial 15 MCQs

SORT & DeepSORT

Drill: What does the Hungarian algorithm assign between frames?

Tutorial 15 MCQs

3D vision

3D vision introduction

Drill: Relate disparity, baseline, and depth with a proportionality sketch.

Tutorial 15 MCQs

Camera calibration

Drill: List intrinsics vs extrinsics; what does reprojection error tell you?

Tutorial 15 MCQs

Stereo vision

Drill: Why does rectification turn matching into a 1D search?

Tutorial 15 MCQs

SLAM

Drill: Name two causes of drift and one mechanism that reduces it.

Tutorial 15 MCQs

CNNs & architectures

CNNs for vision

Drill: Explain parameter sharing and how stride affects output map size.

Tutorial 15 MCQs

AlexNet

Drill: List three design choices that helped 2012 ImageNet scale.

Tutorial 15 MCQs

ResNet

Drill: Write the residual mapping F(x)+x and when projection shortcuts are needed.

Tutorial 15 MCQs

MobileNet

Drill: Contrast depthwise vs pointwise conv in terms of mixing channels vs space.

Tutorial 15 MCQs

Autoencoders

Drill: What does a narrow bottleneck encourage in the latent code?

Tutorial 15 MCQs

Generative models

GANs

Drill: State the minimax roles of G and D; define mode collapse.

Tutorial 15 MCQs

Diffusion models

Drill: Forward noising vs learned reverse step in plain language.

Tutorial 15 MCQs

Video & motion

Video processing

Drill: Why is temporal coherence important for action vs independent frame classification?

Tutorial 15 MCQs

Optical flow

Drill: Write brightness constancy linearized; name the aperture problem.

Tutorial 15 MCQs

Action recognition

Drill: Compare two-stream RGB+flow vs 3D conv on a short clip.

Tutorial 15 MCQs

Applications

Face recognition

Drill: Contrast verification 1:1 vs identification 1:N; name one metric pair.

Tutorial 15 MCQs

Pose estimation

Drill: Top-down vs bottom-up multi-person strategies in two bullets.

Tutorial 15 MCQs

OCR

Drill: Split detection vs recognition; when is CTC used?

Tutorial 15 MCQs

Autonomous vehicles

Drill: List two vision outputs a planner might consume besides raw pixels.

Tutorial 15 MCQs

Tools, metrics & datasets

OpenCV

Drill: Why does OpenCV default to BGR; one fix before matplotlib?

Tutorial 15 MCQs

PyTorch Vision

Drill: What do transforms.Normalize parameters refer to statistically?

Tutorial 15 MCQs

TensorFlow Vision

Drill: Name two tf.data ops that improve GPU utilization.

Tutorial 15 MCQs

CV evaluation metrics

Drill: Relate precision, recall, and AP in one short chain.

Tutorial 15 MCQs

ImageNet

Drill: What is a WordNet synset in the context of class labels?

Tutorial 15 MCQs

COCO dataset

Drill: Instance mask AP vs box AP—what extra alignment is scored?

Tutorial 15 MCQs

Frequently asked questions

Work topic-by-topic: attempt the drill, read the tutorial section you are weakest on, then take the MCQ. Repeat missed topics after a few days.

MCQs include explanations. Short drills are self-checked against the linked tutorial and your own sketches or code snippets.