computer-vision

rohitg00/ai-engineering-from-scratch

A 542-lesson curriculum for building AI systems from the ground up — A free, MIT-licensed course with 542 lessons across 20 phases covering model internals, retrieval pipelines, and agent runtimes in Python, TypeScript, Rust, and Julia. Learners write code, run evals, and inspect failures, with labs and project challenges for self-paced or agent-assisted study.

roboflow/supervision

Model-agnostic building blocks for computer vision pipelines — Supervision offers reusable Python components for detection, segmentation, tracking and annotation that plug into Ultralytics, Transformers, MMDetection or Inference. It's aimed at developers who want to skip boilerplate and focus on building CV applications rather than reimplementing common tooling.

screenpipe/screenpipe

Screenpipe records your work screen to surface automation targets — Screenpipe continuously captures screen and audio on company computers, then maps workflows to find tasks worth automating and feeds context to AI agents. Backed by YC S26, it's built local-first and privacy-conscious, aimed at teams wanting to understand and automate real work patterns rather than guess at them.

kornia/kornia

Kornia aims to be the spec for differentiable vision in PyTorch — Kornia is a PyTorch-based differentiable computer vision library offering batch, GPU-accelerated, auto-differentiable image processing and geometric algorithms. Its stated direction is to become the reference implementation and executable specification for differentiable CV and geometry in the PyTorch ecosystem, useful for researchers and engineers building spatial AI, augmentation pipelines or robotics vision.

rerun-io/rerun

A columnar data layer built for multimodal robotics logging — Rerun ingests multi-rate robotics data—images, point clouds, transforms, time series, video—from logs, sim, or MCAP/LeRobot formats, then lets you visualize it synchronously in a built-in viewer, query it as dataframes or SQL, and stream it into training. Built in Rust with Python, Rust, and C++ SDKs, it targets robotics and physical AI teams handling heterogeneous sensor data.

roboflow/rf-detr

Transformer-based detection with NAS-tuned model sizes — RF-DETR adds keypoint detection (preview) alongside object detection and instance segmentation, built on a DINOv2/PE-Core-T backbone. Its model sizes come from neural architecture search, now also available on Roboflow's platform to find optimal architectures for custom datasets. Useful for developers fine-tuning real-time detectors for their own data.

koharu-rs/koharu

A local-first manga translator built in Rust — Koharu runs detection, OCR, inpainting and LLM translation entirely on-device, keeping manga pages private. It handles raster images, archives and PDFs with page sequencing, and lets users run the pipeline selectively per page or project. Useful for scanlators and hobbyists wanting offline, GPU-accelerated translation with WebGPU.

harry7557558/spirula-studio

A self-contained 3D Gaussian Splatting pipeline that runs without Python — Spirula Studio takes photos or video straight to a textured 3DGS mesh in one binary, with no Python/PyTorch or separate COLMAP install. It runs on NVIDIA, AMD, Intel, and Apple GPUs via Vulkan, fits 10M SH3 Gaussians in 8GB VRAM, and natively handles fisheye/360° cameras, useful for artists and researchers wanting a lightweight, cross-vendor trainer.

withoutbg/withoutbg-python

Python SDK for local and cloud background removal (pip install withoutbg)

NVlabs/alpasim

NVIDIA's Modular Closed-Loop Simulator for End-to-End AV Research — AlpaSim is an open-source, Python-based closed-loop simulator from NVlabs for testing end-to-end autonomous driving policies. It pairs pluggable sensor rendering (NuRec, plus video-model rendering via OmniDreams/FlashDreams) with a microservices architecture, letting researchers swap components and scale across nodes for algorithm validation and benchmarking.