Computer Vision Mastery Path¶

Mastery Path

BOOTCAMP

VINOD BHANUJA

Most computer vision tutorials either stay at "here's a filter" forever, or jump straight to a pretrained model with no idea what it's actually doing underneath. This course builds the whole real stack: classical OpenCV operations run against real images, a real CNN trained from scratch and watched as it actually learns, real pretrained detectors and segmentation models run and evaluated locally, real optical flow and tracking on real video frames, real OCR against real text, and real vision transformer and embedding-based retrieval, all run locally — no cloud vision API, no account, no managed service standing in for understanding.

This course was written and run against a real local Python vision stack — OpenCV, scikit-image, PyTorch and torchvision — with every pixel, gradient, bounding box, mask and tracked point computed on this machine, not asserted from documentation. Where a real result was honest but unglamorous (a detector that missed, a tracker that drifted, a metric that didn't move), it's reported exactly as measured.

It assumes basic Python comfort and does not re-teach general machine learning fundamentals (training loops, loss functions, optimizers) — for that foundation, see the AI/ML Mastery Path. It also does not cover deploying vision models to microcontrollers or edge accelerators (TFLite Micro, NPUs, ESP32-CAM) — for that, see the Edge AI Mastery Path; this course stays at normal desktop/cloud-scale compute throughout.

How the program is organized¶

Level	Focus	Modules	What you can do after
1 · Image Fundamentals & Classical CV	Pixels, color spaces, filtering, contours, features, classical detection	10	Manipulate and analyze real images with OpenCV directly
2 · CNNs & Image Classification	Convolution, training a real classifier, transfer learning, evaluation	10	Train and evaluate a real image classifier end to end
3 · Detection, Segmentation & Motion	Object detection, segmentation, pose, optical flow, tracking, OCR	10	Run and evaluate real detection/segmentation/tracking models
4 · Modern Architectures & Production CV	Vision transformers, embeddings, generation, robustness, bias, scaling	10	Choose, evaluate and ship a real, modern vision system

Each level has 9 teaching modules plus a hands-on project (module 10), for 40 lessons total.

How to use this site¶

Run it yourself. A real local Python environment with opencv-python, scikit-image, torch and torchvision is all this course's core lessons need — no account, no cloud bill, no managed vision API.
Read the "What was run" note at the top of each lesson. The overwhelming majority of this course's images, metrics, and detections come from real, observed runs against real pixels; anything that genuinely couldn't be run locally (massive-scale pretraining, specialized hardware) says so plainly.
Do the "predict, then run" exercises. Vision has several behaviors (what a filter kernel actually does to an edge, where IoU breaks down, how a tracker drifts) that are easy to miss from reading alone and genuinely clarifying to watch happen on a real image.
Do the projects. Each one is a real, runnable pipeline against real images or video, not a diagram.