AI Roadmaps
← All roadmaps
IntermediateRoadmap 08 · 12 steps · 4h 40m of video

Computer Vision

Teach machines to see: CNNs, transfer learning, YOLO detection, segmentation, ViT, and CLIP.

Before this: Deep Learning.

  1. 1h 03m
    Step 1

    Deep learning for vision

    Stanford CS231n lecture 1: the field in one hour.

    Stanford CS231N Deep Learning for Computer Vision | Spring 2025 | Lecture 1: Introduction

    Stanford Online

  2. 9 min
    Step 2

    CNNs explained

    Filters, feature maps, and pooling.

    Convolutional Neural Networks (CNNs) explained

    deeplizard

  3. 22 min
    Step 3

    Build a CNN in PyTorch

    Train an image classifier from scratch.

    PyTorch Tutorial 14 - Convolutional Neural Network (CNN)

    Patrick Loeber

  4. 1h 11m
    Step 4

    CNN architectures

    AlexNet to ResNet: what each one changed.

    Stanford CS231N Deep Learning for Computer Vision | Spring 2025 | Lecture 6: CNN Architectures

    Stanford Online

  5. 15 min
    Step 5

    Transfer learning

    Reuse a pretrained network and train on a few hundred images.

    PyTorch Tutorial 15 - Transfer Learning

    Patrick Loeber

    huggingface/pytorch-image-models★ 37.3k
  6. 3 min
    Step 6

    Data augmentation

    Get more out of small datasets for free.

    Data Augmentation explained

    deeplizard

    roboflow/notebooks★ 9.7k
  7. 17 min
    Step 7

    How YOLO works

    Real-time object detection in one pass.

    How YOLO Object Detection Works

    DeepBean

  8. 21 min
    Step 8

    Train YOLO on your own data

    Label, train, and run a detector in Colab.

    How to Train YOLO Object Detection Models in Google Colab (YOLO26, YOLO11, YOLOv8)

    Edje Electronics

    ultralytics/ultralytics★ 63.6k
  9. 11 min
    Step 9

    U-Net segmentation

    Classify every pixel, the architecture diffusion models also use.

    The U-Net (actually) explained in 10 minutes

    rupert ai

  10. 13 min
    Step 10

    Segment Anything

    A foundation model that segments any object from a click.

    Explaining the Segment Anything Model - Network architecture, Dataset, Training

    Neural Breakdown with AVB

    facebookresearch/sam2★ 20.2k
  11. 17 min
    Step 11

    Vision transformers

    Images as sequences of patches, with code.

    Vision Transformer Quick Guide - Theory and Code in (almost) 15 min

    DeepFindr

  12. 18 min
    Step 12

    CLIP

    Connect images and text in one embedding space.

    How AI 'Understands' Images (CLIP) - Computerphile

    Computerphile

    openai/CLIP★ 34.6k

Build it

Watching isn't learning. Pick one and put it on GitHub.

Where to go next