Short, concrete theory
One idea at a time, with the formula, the numbers and the pictures. Built from a university course on computer vision.
Teach a computer to see, one algorithm at a time. Every lesson runs real OpenCV programs in Python right in your browser, shows the same program in C++, and lets you take each algorithm apart in a live image lab.
Python runs in your browser · progress saved locally
Vision is learned by watching algorithms work on real images. Each lesson alternates short explanations with programs you run and labs where you move every parameter.
One idea at a time, with the formula, the numbers and the pictures. Built from a university course on computer vision.
Press Run and the Python program executes in your browser with OpenCV 4.10. Edit it, script key presses, see every window it opens.
Every program exists in both languages with the same windows and keys. Download either file and run it on your machine.
Drag sliders on real images, watch histograms, kernels and RANSAC update, then check yourself with an explained quiz and a practice task.
Both run in this page. Press Run on the Python tab, change a threshold, run again — or switch to C++ to see the same program built with g++.
Lesson 5 · the Canny edge detector
Lesson 13 · features, RANSAC and a homography
Five parts, each answering a question the previous part leaves open. Start at lesson 1 — every lesson assumes only the ones before it, plus basic Python.
When you finish lesson 14 you will not just know which OpenCV function to call. You will know what it computes, which parameter breaks it, and how to write it yourself.
Point operations, histograms, thresholds, convolution kernels, edges and morphology — and a change detector that finds people in a surveillance video.
The pinhole model x = K[R | t]X, lens distortion, calibration from a chessboard and camera pose from known points.
Harris, SIFT and ORB features, ratio-test matching, optical flow, least squares and RANSAC — ending with a panorama stitcher.
Homographies, triangulation, two-view geometry, structure-from-motion and the volume rendering behind NeRF.
“Graphics turns a 3D world into pixels. Vision runs the film backwards — and every lesson in this course is one frame of that reversal.”
Computer Vision I · lesson 1The practical details.
pip install opencv-python numpy scipy matplotlib for Python, and OpenCV 4 with g++ -std=c++17 file.cpp $(pkg-config --cflags --libs opencv4) for C++.imshow window becomes an image card under the code; video loops become a frame scrubber. waitKey() reads from a Keys box, so you can script +++ or {tab} presses, and mouse callbacks are driven by a scripted mouse path. On your machine the same programs open real windows.Then fourteen lessons take you from pixels and filters through camera geometry and feature matching to 3D reconstruction. One idea, one program, one lab at a time.
Free · No sign-up · Real OpenCV output