Environment:Kornia Kornia PyTorch Python Environment

Knowledge Sources	Kornia PyPI kornia
Domains	Infrastructure, Computer_Vision
Last Updated	2026-02-09 15:00 GMT

Overview

Python 3.11+ environment with PyTorch >= 2.0.0 and kornia_rs >= 0.1.9 native backend for differentiable computer vision.

Description

This is the core runtime environment required to use the Kornia library. Kornia is built entirely on top of PyTorch and requires a modern Python runtime (3.11 or newer). The library uses a Rust-based native extension (kornia_rs) for high-performance image I/O operations using libjpeg-turbo and other native codecs. The packaging library is also required for version comparison utilities used internally by Kornia's compatibility layer.

Usage

Use this environment for any Kornia operation. This is the base prerequisite for all Kornia workflows including image augmentation, feature matching, edge detection, and ONNX pipelines. Every Implementation page in the Kornia wiki requires this environment.

System Requirements

Category	Requirement	Notes
OS	Linux, macOS, Windows	Cross-platform; `kornia_rs` provides native binaries per OS
Hardware	CPU (minimum)	GPU optional; see CUDA_GPU_Environment for GPU acceleration
Disk	~500MB	For PyTorch + Kornia + kornia_rs installation

Dependencies

System Packages

No special system packages required for CPU-only usage
For `kornia_rs` image I/O: libjpeg-turbo is bundled in the wheel

Python Packages

`python` >= 3.11 (supports 3.11, 3.12, 3.13)
`torch` >= 2.0.0
`kornia_rs` >= 0.1.9
`packaging` (any version)

Credentials

No credentials are required for core Kornia functionality.

Quick Install

# Install Kornia (automatically pulls torch, kornia_rs, packaging)
pip install kornia

# Or install with specific PyTorch version
pip install torch>=2.0.0 kornia

Code Evidence

Python version requirement from `pyproject.toml:9`:

requires-python = ">=3.11"

Core dependencies from `pyproject.toml:32`:

dependencies = ["kornia_rs>=0.1.9", "packaging", "torch>=2.0.0"]

PyTorch version comparison utilities from `kornia/core/_compat.py:74-102`:

def torch_version_ge(major: int, minor: int, patch: Optional[int] = None) -> bool:
    """Check if the current PyTorch version is greater than or equal to the specified version."""
    _version = version.parse(torch_version())
    if patch is None:
        return _version >= version.parse(f"{major}.{minor}")
    else:
        return _version >= version.parse(f"{major}.{minor}.{patch}")

kornia_rs is imported directly (not lazy-loaded) for image I/O from `kornia/io/io.py:24`:

import kornia_rs

Common Errors

Error Message	Cause	Solution
`ModuleNotFoundError: No module named 'kornia_rs'`	kornia_rs native extension not installed	`pip install kornia_rs>=0.1.9`
`RuntimeError: Dynamo is not supported on Python 3.12+`	torch.compile limitation on newer Python	Downgrade to Python 3.11 or disable torch.compile
`ImportError: torch`	PyTorch not installed	`pip install torch>=2.0.0`

Compatibility Notes

Python 3.11+: Required minimum. Python 3.10 and earlier are not supported.
PyTorch JIT Scripting: Kornia uses `Optional[]`, `Union[]`, `List[]` typing syntax (not `X | None`) for compatibility with `torch.jit.script`. This is documented in `pyproject.toml` ruff ignore rules `UP006`, `UP007`, `UP035`.
torch.compile: Supported but may have issues on Python 3.12+. Kornia avoids in-place operations in critical paths to maintain `torch.compile()` compatibility.
MPS (Apple Silicon): Supported with fallbacks. Some operations like `torch.std_mean` have MPS-specific workarounds via `is_mps_tensor_safe()`.

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment