How andandandand/practical-computer-vision works
Project type and purpose
• This is an educational/tutorial repository (“Practical Computer Vision with PyTorch and FiftyOne”). It provides Jupyter notebooks, data-downloading scripts, documentation, and slide decks to teach computer-vision workflows: data ingestion, preprocessing, model training, embedding extraction, visualization, and evaluation.
File structure
Top-level directories:
• .devcontainer/ – VS Code Dev Container configuration
• artifacts/ – pre-downloaded datasets, embeddings, class mappings
• docs/ – step-by-step setup guides (Python install, Kaggle, wandb, etc.)
• image_scraping/ – scripts & notebooks for scraping images from Bing/Google Drive
• notebooks/ – core tutorial notebooks covering digital images, CNNs, U-Net, CLIP, diffusion, etc.
• slides/ – workshop slide decks for computer-vision series
• src/data_downloaders/ – Python modules to programmatically fetch FiftyOne datasets
• requirements.txt – pinned Python dependencies
README for additional insights
• The README enumerates: setup steps, tutorial notebook list, external integrations (Kaggle, wandb, HuggingFace, Google Drive), and teaching objectives (embeddings, classification, segmentation, diffusion, interpretability).
• Technology stack: Python 3.11, PyTorch, FiftyOne, Docker (Dev Container), Kaggle/Colab, wandb, HuggingFace.
Essential architectural elements
• Do not overcrowd: group similar notebooks into a single “Tutorial Notebooks” module, but list key sub-domains (classification, segmentation, embeddings).
• How the Dev Container ensures consistent dependencies.
• Offline artifacts folder as the central data “cache” across tutorials.
• External dependencies clearly to show users what credentials/API tokens they need (HuggingFace, wandb, Google Drive).