Skip to content
Datamewby Intura
Start free

Lessons

Computer Vision with Deep Learning: all lessons

Teaching machines to see: from pixels to object detection.

6 modules · 17 lessons · 1 free to read now

About this track: outcomes, tools and price

Module 1Image Fundamentals

3 lessons

Images as numbers, convolution as a sliding weighted sum, and why a CNN learns its own filters.

  1. How Computers See Images

    18 min

    To a computer an image is a table of numbers. Convolution — a small weighted sum slid across it — turns those numbers into edges, textures and eventually objects.

    Free · read nowRead the lesson Simulation inside

  2. Image Preprocessing and Augmentation

    16 min

    Resizing, normalisation and augmentation that make a model robust to the real world.

    Included with the track Simulation inside

  3. Pooling and Receptive Fields

    14 min

    How deeper layers come to "see" larger parts of the image.

    Included with the track Simulation inside

Module 2Image Classification with CNNs

3 lessons

Building a CNN in PyTorch, transfer learning from pretrained models, and evaluating a classifier honestly.

  1. Building a CNN in PyTorch

    22 min

    Conv2d, pooling and a classification head — from scratch to trained.

    Included with the track Simulation inside

  2. Transfer Learning with Pretrained Backbones

    18 min

    Freezing, fine-tuning, and different learning rates per layer.

    Included with the track Simulation inside

  3. Evaluating Image Classifiers

    16 min

    Confusion matrices, imbalanced classes, and looking at the mistakes.

    Included with the track Simulation inside

Module 3Object Detection

3 lessons

Bounding boxes, IoU, one-stage detectors like YOLO, and how to read mAP.

  1. Bounding Boxes and IoU

    15 min

    How to measure how good a box is, and non-maximum suppression.

    Included with the track Simulation inside

  2. One-Stage Detectors: YOLO

    18 min

    Predicting classes and boxes together, in one forward pass.

    Included with the track Simulation inside

  3. Training a Custom Detector

    20 min

    Annotation, label formats and fine-tuning on your own objects.

    Included with the track Simulation inside

Module 4Image Segmentation

3 lessons

Classification per pixel: semantic and instance segmentation, U-Net, and general segmentation models.

  1. Semantic vs Instance Segmentation

    14 min

    Labels per pixel, masks per object, and the metrics for each.

    Included with the track Simulation inside

  2. U-Net

    18 min

    The encoder, the decoder, and the skip connections that bring back detail.

    Included with the track Simulation inside

  3. Promptable Segmentation Models

    14 min

    General models that segment anything from a click or a box.

    Included with the track

Module 5Vision Transformers and Multimodal Models

3 lessons

ViT, CLIP and vision-language models — when images and text share one embedding space.

  1. Vision Transformers (ViT)

    18 min

    Patches as tokens, and when a ViT beats a CNN.

    Included with the track Simulation inside

  2. CLIP and Image-Text Embeddings

    18 min

    Zero-shot classification and searching images with a sentence.

    Included with the track Simulation inside

  3. Vision-Language Models

    16 min

    LLMs that read images: what they can do, what they cannot yet, and how to test them.

    Included with the track

Module 6Computer Vision in Production

2 lessons

Models on edge and mobile devices, real-time camera latency, and a capstone project.

  1. Edge and Mobile Deployment

    16 min

    Small models, quantisation and runtimes for phones and cameras.

    Included with the track Simulation inside

  2. Capstone: a Vision System

    1h

    From an annotated dataset to a running, evaluated detector.

    Included with the track Simulation inside