Learn computer vision

Computer Vision Fundamentals

48 articles so far, ordered by dependency: nothing appears before the thing it depends on. Greyed units are planned and not written yet.

F1

Imaging

How light becomes an array of numbers.

1.1 Image formation

How a scene is flattened onto a sensor, and what the glass does on the way.

4 lessons · unit overview

1. 
   1 The pinhole model Beginner 8 min
2. 
   2 Lenses and depth of field Beginner 8 min
3. 
   3 Lens distortion Intermediate 9 min
4. 
   4 Wide-angle and fisheye Intermediate 9 min

1.2 Image sensing

What the number in a pixel counts, and how far it can be trusted.

5 lessons · unit overview

1. 
   1 From photons to counts Intermediate 9 min
2. 
   2 Noise and dynamic range Intermediate 9 min
3. 
   3 The response curve Intermediate 8 min
4. 
   4 High dynamic range Intermediate 9 min
5. 
   5 Sensing colour: the Bayer CFA Intermediate 6 min

1.3 The image as data

What the array holds, where its samples sit, how finely they are written, and what the file did to them.

6 lessons · unit overview

Before this: Image sensing

1. 
   1 An image is an array Beginner 7 min
2. 
   2 Sampling: the pixel grid Beginner 9 min
3. 
   3 Quantization and bit depth Beginner 7 min
4. 
   4 Colour spaces Beginner 9 min
5. 
   5 How close is your colour? Intermediate 8 min
6. 
   6 File formats and compression Beginner 9 min

F2

Image processing

What you do to that array once you have it.

2.1 Spatial filtering

The one operation behind blur, sharpening, edges and every CNN.

1 lesson

Before this: An image is an array

1. 
   1 How 2D convolution works Beginner 8 min

2.2 Binary images Planned Thresholding, morphology, and measuring the blobs that are left.

2.3 The frequency domain Planned The Fourier transform of an image, and why you blur before you downsample.

F3

Features

What in an image is worth pointing at, and how to describe it.

3.1 Edge detection

Finding where one thing stops and another begins.

5 lessons · unit overview

Before this: 2D convolution

1. 
   1 What is an edge? Beginner 9 min
2. 
   2 Image gradients Intermediate 10 min
3. 
   3 Canny, taken apart Intermediate 9 min
4. 
   4 Laplacian, LoG and zero-crossings Intermediate 9 min
5. 
   5 Corners: the structure tensor Intermediate 10 min

3.2 Boundary detection

Finding lines and circles by voting: the Hough transform, and fitting one line to what it finds.

4 lessons · unit overview

Before this: Edge detection

1. 
   1 The Hough transform Intermediate 8 min
2. 
   2 The probabilistic Hough transform Intermediate 8 min
3. 
   3 From segments to one line Intermediate 8 min
4. 
   4 The Hough circle transform Intermediate 14 min

3.3 SIFT

Finding the same point in two photos: scale space, keypoints, 128 numbers per point, and matches you can trust.

5 lessons · unit overview

Before this: Edge detection

1. 
   1 What makes a good keypoint? Intermediate 14 min
2. 
   2 Scale space and the DoG Intermediate 14 min
3. 
   3 How SIFT finds keypoints Intermediate 14 min
4. 
   4 The SIFT descriptor Intermediate 12 min
5. 
   5 Lowe's ratio test Intermediate 10 min

3.4 Beyond SIFT Planned Binary descriptors, and detectors and matchers that are learned.

3.5 Image alignment

Mapping one view of a plane onto another: the 3×3, how to fit it, and how to warp through it.

5 lessons · unit overview

Before this: Homogeneous coordinates

1. 
   1 Rotate, scale, shear Beginner 7 min
2. 
   2 Affine vs perspective Beginner 8 min
3. 
   3 Compute a homography Intermediate 9 min
4. 
   4 RANSAC Intermediate 16 min
5. 
   5 Bird’s-eye view and panoramas Intermediate 9 min

F4

Geometry

From one camera to many, and back out to three dimensions.

4.1 Camera geometry

One matrix from a 3-D point to a pixel, and running it backwards: calibration and pose.

4 lessons · unit overview

Before this: Image formation

1. 
   1 Homogeneous coordinates Intermediate 8 min
2. 
   2 The camera matrix Intermediate 5 min
3. 
   3 Calibration Intermediate 5 min
4. 
   4 PnP: pose from known points Intermediate 5 min

4.2 Two views Planned Epipolar geometry, triangulation, and depth from a stereo pair.

4.3 Many views

Recovering structure and motion from a set of photographs.

1 lesson

Before this: SIFT, RANSAC, PnP

1. 
   1 Structure from motion Advanced 16 min

4.4 Motion Planned Optical flow: brightness constancy, the aperture problem, Lucas–Kanade.

OpenCV in Practice

Build one real project per module with Python and OpenCV, running every line in your browser, and see where it works and where it fails. Each module links the Fundamentals units that explain why it works. The code runs in the page; the first run downloads about 20 MB.

M1 Fix a badly exposed photo

Open an image, read its pixels and histogram, and fix its exposure, measured against a longer exposure of the same scene.

6 lessons · project overview

Why it works: Unit 1.2: Image sensing, Unit 1.3: The image as data

1. 1 Open an image and read its pixels 3 min
2. 2 Image histograms 3 min
3. 3 Brightness and contrast 3 min
4. 4 Gamma correction 3 min
5. 5 Histogram equalization and CLAHE 3 min
6. 6 Crop, resize and save 3 min

M2 Find every ball on a pool table

Colour spaces, masks, morphology and contours: pick out objects by colour, then count what the pipeline gets right and wrong.

5 lessons · project overview

Why it works: Unit 1.3: The image as data

1. 1 Find the HSV range of an object 4 min
2. 2 Colour masks with inRange 3 min
3. 3 Erode, dilate, open and close 4 min
4. 4 Find and filter contours 4 min
5. 5 Where colour detection fails 6 min

The Computer Vision Metrics Handbook

Every score in computer vision, from the ground up. A self-contained book on evaluation with its own reading order: one confusion matrix grows into precision and recall, and those stretch over boxes, masks, identities through time, and images with no ground truth at all.

The handbook has its own page ›

1. 01

   Classification

   Classification Metrics, From the Ground Up: Precision, Recall, F1, ROC and AUC

   The confusion matrix and everything that falls out of it: precision, recall, F1, specificity, the precision/recall trade-off, PR and ROC curves, and AUC, all built on one running toy example (a backyard cat-cam), with the equations, the intuition, and scikit-learn.

   Confusion matrixPrecisionRecallF1SpecificityPR & ROC curvesAUC
2. 02

   Multiclass & multilabel

   Multiclass and Multilabel Metrics: Micro, Macro, Weighted, and the Difference That Matters

   How precision/recall/F1 scale past two classes: one-vs-rest, and the micro/macro/weighted averaging that decides whether rare classes count. Plus multilabel-specific metrics: subset accuracy, Hamming loss, and top-k, all on one running shelter-camera example, with scikit-learn.

   One-vs-restMicroMacroWeightedSubset accuracyHamming lossTop-k
3. 03

   Object detection

   Object Detection Metrics: IoU, AP, and the mAP Everyone Quotes

   How detection is scored: IoU to decide a hit, greedy matching into TP/FP/FN, per-class Average Precision as the area under the precision–recall curve, and the COCO mAP@[.5:.95] you see in every paper, all on one running parking-lot example, with torchmetrics.

   IoUAPPR curvemAP
4. 04

   Segmentation

   Segmentation Metrics: IoU, Dice, mIoU, and Panoptic Quality

   Scoring masks instead of boxes: pixel IoU/Jaccard and the Dice coefficient (and why they're related), mIoU for semantic segmentation, mask-AP for instance segmentation, and Panoptic Quality for the unified task, on one running cat-mask example, with the equations and code.

   IoU / JaccardDicemIoUMask APPanoptic Quality
5. 05

   Object tracking

   Object Tracking Metrics: MOTA, MOTP, IDF1, and HOTA

   Tracking adds identity over time, so its metrics score two things at once: detecting objects and keeping their IDs consistent. MOTA, MOTP, ID switches, IDF1, and the modern HOTA that balances detection against association, on one running door-cam example, with the equations and motmetrics.

   MOTAMOTPID switchesIDF1HOTA
6. 06

   Image generation

   Image Generation Metrics: FID, IS, KID, LPIPS, and CLIPScore

   How do you score images with no ground truth? You compare distributions and use learned perceptual models: Inception Score, Fréchet Inception Distance, KID, LPIPS, and CLIPScore for prompt fidelity, on one running cat-generator example, with the equations, the catches, and torchmetrics.

   Inception ScoreFIDKIDLPIPSCLIPScore