kornia-arch
Skill748 starsChanged 7 days agoApache-2.0View on GitHub
Use first when learning or navigating kornia-rs — workspace layout, crate responsibilities, feature flags, and core types (Tensor, Image<T, C, A>). Trigger on "where does X live", "which crate", "how is the workspace organised", or before touching an unfamiliar crate.
Files
kornia-rs architecture
Low-level 3D computer vision library in Rust (edition 2021, MSRV 1.89), with PyO3/Maturin Python bindings and C++ bindings.
Layout
crates/*— core library crates (workspace default members)examples/*— standalone example binarieskornia-py/— Python bindings (PyO3/Maturin,python/,tests/)kornia-cpp/— C++ bindings (CMake,include/,tests/)scripts/— release/version helpers (release_rust.sh,check_version_pins.py)
Crates
| Crate | Purpose |
kornia-tensor | N-dimensional tensor |
kornia-tensor-ops | Tensor operations |
kornia-image | Image<T, C, Alloc> built on tensors (ImageError) |
kornia-imgproc | resize, color, filters, warp, features, SIMD kernels |
kornia-io | image/video I/O (turbojpeg, gstreamer, v4l) |
kornia-3d | ICP, point clouds, stereo |
kornia-algebra | linear algebra |
kornia-calib | multi-camera extrinsic calibration (AprilTag BA) |
kornia-apriltag, kornia-bow, kornia-vlm | detection, bag of words, VLM (candle) |
kornia | umbrella crate that re-exports the others |
The publish set is PUBLISH_ORDER / NOT_PUBLISHED in scripts/release_rust.sh
(crates with publish = false are skipped). Read it there; do not restate it.
Feature flags
ci = arrow, bincode, gstreamer, serde, turbojpeg (needs nasm),
v4l (needs clang). cuda is not in ci (needs hardware). OpenCV
comparison code in kornia-imgproc is behind opencv_bench and is not built
by the lint workflow.
Key patterns
Image<T, C, Alloc>:Tpixel type (u8,f32),Cchannel count.- imgproc ops take
&srcand a pre-allocated&mut dst. - I/O entry points live in
kornia::io::functional. - Calibration thresholds (
max_reprojection_error,robust_scale_sq) are in normalized image coordinates, not pixels. - Linear algebra in new code (kornia-rs#542): vector/matrix types from
kornia-algebra(glam-backed), decompositions fromcrates/kornia-algebra/src/linalg/(SVD, Cholesky, rigid), other solvers on nalgebra. No hand-rolled inverses, eigen-solvers ordot/crosshelpers.
See also: kornia-workflow, kornia-coding-style, kornia-api-flavor.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- kornia-api-flavorSkill
- kornia-bindingsSkill
- kornia-code-reviewSkill
- kornia-coding-styleSkill
- kornia-curationSkill
- kornia-simdSkill
- kornia-workflowSkill
- releaseSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.