Dev Playground

Simple containerized environment to quickly try random stuff, with a focus on python and deep learning tools.

Also includes some experiments and results.

Features

• Dockerfile:
  • ubuntu24.04
  • nvidia-cuda12.6
  • cpp tools
  • python tools
  • node/js tools
  • shell tools
  • xauth/X11
  • miniconda
• vscode configs (vscode/settings.json and devcontainer.json)
  • gpu support
  • graphical interface support (X11)
  • host camera support
• Pre-commit hooks
• Python tool configs (linters, formatters)
• Node/js tool configs (linters, formatters)
• CI/CD (github actions) for formatting/linting

---

Explored topics

• devcontainer: Containerized docker environment with dev tools and configs.

• diffusion: Data generation using DDPMs (conditional and unconditional generation).

• dataset_creator: CLIP based text and image search over an image directory and autolabeling using foundational models.

• traffic_detection: Traffic analysis using computer vision to detect, track and count vehicles from traffic cameras.

• industrial_detection: Use pretrained models to solve industrial robot tasks.

• llm_app: LLMs + Retrieval Augmented Generation (RAG) with a Web chat interface (similar to openAI but worse).

• dataset_selection: Selecting of most diverse images of a directory based on similarity algorithm (classical CV).