TodayTools

Computer vision flashcards

Teach computers to see: classify, detect and segment what is in a picture.

Donate
Beginner friendly, starts from zero
Section
How to study
0
Computer visionStart from zero · 1 of 11

What is computer vision, in one plain sentence?

Teaching computers to understand pictures and video: what is in them, where it is and what is happening, the way your eyes and brain do without effort.

Cameras that understand, not just record.
Show again
11 left
Got it

Tap to turn over. Swipe right if you know it, left to see it again.

All topics

What you will learn

  1. Start from zero

    • What computer vision is2 cards
    • How a computer sees2 cards
    • The main tasks3 cards
    • How models learn2 cards
    • Limits and care2 cards

  2. Core basics

    • Images as arrays3 cards
    • Preparing images3 cards
    • Neural networks for images4 cards
    • Training basics3 cards
    • Measuring results5 cards

  3. Next steps

    • Object detection5 cards
    • Segmentation3 cards
    • Custom data4 cards
    • Text and people2 cards
    • Video2 cards
    • Deployment2 cards

  4. Go deeper

    • Modern architectures6 cards
    • Training pitfalls4 cards
    • Edge deployment3 cards
    • Robustness and security2 cards
    • Responsible use2 cards
    • Beyond 2D2 cards

Final test and certificate

10 questions drawn fresh from the whole topic, about 15 minutes. Sign in, score 80% or more and get a free certificate with your name and a public check link anyone can open, such as an employer or a teacher. Not signed in? Try a practice test from the cards.

Take the final test

About these cards

  • Reviewed 26 October 2026
  • CoversOpenCV 4.xPyTorch 2.xtorchvision 0.2xPython 3.13
How these cards are made

Computer vision at a glance

Computer vision is how computers find meaning in pictures and video, such as spotting a face, reading a number plate or finding a crack in a pipe.

Like teaching a child to tell cats from dogs: you show many labelled pictures until they can name a new one they have never seen.

Words you will meet

Pixel
One tiny coloured dot; a photo is a grid of millions of them
Image classification
Naming what the whole picture shows, like "cat"
Object detection
Finding each object and drawing a box around it
Bounding box
The rectangle drawn around a found object
Segmentation
Colouring the exact pixels that belong to each object
Label
The correct answer a person attached to a training picture
Model
The program that learned from labelled pictures and now makes guesses
Dataset
The big collection of labelled pictures used to teach and test

The steps

  1. 1. Start from zero (11 cards): What computer vision is, how computers see pictures, the main tasks and the words you need.
  2. 2. Core basics (18 cards): Images as arrays, preprocessing, CNNs, pretrained models, training and metrics.
  3. 3. Next steps (18 cards): Detection, segmentation, custom data, OCR and pose, video and deployment.
  4. 4. Go deeper (19 cards): Transformers and image-text models, training pitfalls, edge speed, robustness and responsible use.
All 11 cards in Start from zero
What is computer vision, in one plain sentence?
Teaching computers to understand pictures and video: what is in them, where it is and what is happening, the way your eyes and brain do without effort.
Where do you meet computer vision every day?
Face unlock on your phone, photo search for "dog", number plate cameras, document scanning apps, self-checkout tills that spot items, and cars that warn when you drift out of lane.
How does a computer "see" a photo?
As a grid of pixels, each one stored as numbers. A colour pixel has three numbers from 0 to 255 for red, green and blue. A phone photo can hold 12 million pixels, so 36 million numbers.
Why is seeing hard for a computer when it is easy for you?
The same cat looks totally different in numbers when the light, angle, size or background changes. The computer must learn which patterns stay the same, from many examples.
What is image classification?
Giving one name to the whole picture, such as "ripe mango" or "not ripe". It answers "what is this?" but not "where is it?".
What are object detection and a bounding box?
Detection finds every object of interest and draws a bounding box, a rectangle, around each, with a name and a confidence score, such as "person 0.94".
What is segmentation?
Colouring the exact pixels of each object instead of a rough box. A photo editor uses it to cut a person out of the background; a farm drone uses it to measure the area of diseased leaves.
What are a label, a dataset and a model?
A label is the correct answer a person gives a picture. A dataset is thousands of labelled pictures. A model is the program that learns from the dataset and then guesses labels for new pictures.
Why do we keep some labelled pictures hidden from the model during learning?
To test it fairly. Like an exam with new questions, pictures it never saw show whether it learned the idea or just memorised the training pictures.
Why can a vision system fail on some people or places?
It only knows what its pictures showed it. If training photos were mostly daylight or mostly one skin tone, it may fail at night or on other people. Good teams test on every group.
What do you need before Core basics, and what will you be able to do?
Basic Python and a first idea of machine learning. By the end you will load images in code, run ready-made models, train on your own photos and measure how good they are.
Computer vision cheatsheet

OpenCV essentials

  • cv2.imread(path) Load image as BGR uint8
  • cv2.cvtColor(img, cv2.COLOR_BGR2RGB) Fix channel order
  • cv2.resize(img, (w, h)) Resize; size is (width, height)
  • cv2.GaussianBlur(img, (5, 5), 0) Smooth noise
  • cv2.Canny(gray, 50, 150) Find edges
  • cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 2) Draw a box

torchvision

  • Weights.DEFAULT Best available pretrained weights
  • weights.transforms() Matching preprocess
  • model.eval() Inference mode for batch norm and dropout
  • torch.no_grad() No gradient tracking
  • torchvision.ops.nms Remove duplicate boxes
  • transforms.v2 Augment images and boxes together

Shapes and formats

  • (H, W, C) NumPy and OpenCV image
  • (N, C, H, W) PyTorch batch
  • XYXY Left, top, right, bottom
  • XYWH Left, top, width, height (COCO)
  • cx, cy, w, h in 0 to 1 YOLO-style labels

Metrics

  • IoU Overlap over union of two boxes
  • precision / recall Right alarms / caught cases
  • mAP@0.5 Detection score at IoU 0.5
  • mAP@[.5:.95] COCO average over IoU thresholds
  • Dice Mask overlap score
  • top-1 / top-5 Right label first / in top five

Task picker

  • classification What is in the image
  • detection What and where, as boxes
  • segmentation Exact pixels per object
  • OCR Read text
  • pose Body keypoints
  • tracking Same object across frames
Donate