π Geometric Computer Vision Library for Spatial AI
Python artificial-intelligencecomputer-visiondeep-learninghacktoberfestimage-processingmachine-learningneural-networkpythonpytorchroboticsspatial-ai
Claude guide for kornia
@AGENTS.md
The operational guide is AGENTS.md; the line above imports it for Claude Code.
When helping a user in this repository:
- Start from the smallest relevant code path and run focused checks before making broad claims.
- Prefer existing kornia utilities and nearby patterns over introducing a parallel abstraction.
- Preserve tensor device, dtype, batch semantics, gradients, and
torch.compilebehavior when they apply. - Treat float16, bfloat16, MPS, CUDA, JIT, and ONNX as distinct compatibility surfaces; do not assume a CPU float32 result covers them.
- Explain what you verified and what remains unverified.
- Ask before gathering broad environment information. Request only useful versions and hardware details, and remind users to redact paths, usernames, tokens, hostnames, and private data names.
If you help prepare a contribution, summarize what you changed and checked so the user can reuse that information in the pull request.