Computer Vision Engineer Roadmap 2026
Build vision systems: image classification, CNNs, detection, segmentation, video, generative models and deployment, following the CS231n arc.
4 stages · 8 units · Advanced · about 150 hours · Last reviewed 19 November 2026
What is a computer vision engineer?
Computer vision is the field of building software that understands images and video. Engineers use it to classify pictures, find and outline objects, track motion, reconstruct 3D scenes and generate images, mostly with neural networks. This roadmap follows the arc of Stanford's CS231n, from pixels and linear classifiers to CNNs, transformers and deployment.
Step-by-step computer vision engineer roadmap
- 1
Foundations
Intermediate · about 38 hours
Images and classifiers.
- Image Basics
Image basics treat a picture as numeric data and build the first classifier on it. An image is an array of pixel values with height, width and colour channels. A linear classifier multiplies the flattened pixels by a weight matrix to produce class scores, trained with a loss such as softmax cross-entropy and gradient descent.
- Neural Networks
Neural networks for vision are layered functions whose weights are learned by backpropagation. The forward pass computes outputs and a loss, and the backward pass applies the chain rule through each layer to get gradients. This unit builds the mechanics you need before convolutions: activations, initialisation, batch normalisation and optimisers.
- Image Basics
- 2
Deep Vision
Advanced · about 38 hours
CNNs and attention.
- CNNs
Convolutional neural networks are networks that slide small learned filters across an image, sharing weights across positions so they detect local patterns such as edges and textures. Stacking convolution, nonlinearity and pooling layers builds up to object-level features. CNNs were the dominant vision architecture for classification before transformers and are still widely used.
- Modern Architectures
Modern architectures are the post-CNN designs now common in vision, chiefly attention-based models such as the Vision Transformer, which splits an image into patches and processes them with self-attention. The unit also covers how large models are trained: transfer learning, mixed precision, distributed training and tuning, which matter as much as the architecture.
- CNNs
- 3
Vision Tasks
Advanced · about 38 hours
Beyond classification.
- Localisation
Localisation tasks ask not only what is in an image but where. Object detection predicts bounding boxes with class labels, using methods such as Faster R-CNN or YOLO. Segmentation labels pixels, either by class or by individual object. Both rely on a CNN or transformer backbone plus task-specific heads and evaluation metrics.
- Temporal and 3D
Temporal and 3D vision extends image models to video and to scene geometry. Video models add a time dimension, using 3D convolutions, two-stream networks or spatiotemporal transformers to recognise actions and track motion. 3D vision recovers depth and structure from images, using stereo, point clouds or learned representations such as neural radiance fields.
- Localisation
- 4
Generative and Production
Advanced · about 38 hours
Create and ship.
- Learning without Labels
Learning without labels covers methods that train vision models using unlabelled images. Self-supervised learning builds a pretext or contrastive task, such as matching two augmented views of one image, to learn useful features. Generative models such as GANs, VAEs and diffusion models learn the data distribution itself and can synthesise new images.
- Deployment
Deployment is the work of moving a trained vision model into a product that runs fast and reliably. It covers exporting to formats like ONNX, shrinking models through quantisation and pruning, serving with batching on a GPU or running on a phone or edge device, and monitoring accuracy once real-world images start arriving.
- Learning without Labels
Frequently asked questions
Ready to practise for interviews?
Practise computer vision engineer questions with an AI mock interviewer and get feedback on your answers.