Training school programme

The programme is organised by topics. Each topic builds on the previous one, guiding participants from foundational skills to advanced applications of computer vision in archaeology.

Programme

Python for Computer Vision Tasks

Setting up a Python environment and Jupyter notebooks
Key libraries: NumPy, Matplotlib, OpenCV
Working with images in Python: reading, displaying, and basic manipulation

Datasets and Annotation

Overview of computer vision tasks: classification, object detection, segmentation
Annotation tools and common formats (COCO, YOLO etc.)
Best practices for creating, managing, and documenting datasets

Data Pre-processing

Image transformations, colour normalisation, and data augmentation
Splitting data and handling class imbalance
Building data pipelines with PyTorch Dataset and DataLoader

Training

Core concepts: CNNs, transfer learning, pre-trained models (ResNet, YOLO)
Setting up training: loss functions, optimisers, and monitoring metrics
Practical session: training a model on an archaeological dataset

Post-processing

Interpreting outputs: confidence scores, bounding boxes, masks
Evaluation metrics: precision, recall, mAP, IoU
Error analysis and visualising results

Advanced Topics

Vision-language models: zero-shot classification, prompt-based detection (CLIP)
Instance and semantic segmentation in detail
Deploying models

Case Studies

Throughout the school, participants will apply the methods above to two real-world archaeological datasets:

Artefact types: classification and detection of archaeological artefacts from photographs
Microscopic use-wear analysis: identifying and classifying wear traces from microscopic images

Keynote

Will be specified…