+
    õäûjÞ  ã                   óH  € R t ^ RIHt ^ RIHt ^RIHt ^RIHt ^RIH	t	H
t
HtHtHtHtHtHtHtHtHtHtHtHt ^RIHt ^RIHt ^ RIHt ^ R	IHt ^R
IHtHtH t H!t!H"t"H#t#H$t$H%t%H&t&H't'H(t(H)t)H*t*H+t+H,t,H-t-H.t.H/t/H0t0H1t1 ]! ]2R4      t3]! ]2R4      t4]! ]2R4      t5R R lt6Rt7R# )uæ  Kornia â€” Differentiable computer vision and image processing for PyTorch.

Kornia's operators work on batched ``torch.Tensor`` inputs, run on the input's device (CPU, CUDA or Apple MPS)
without a NumPy round trip, and most are differentiable. Kornia pays off where a hand-written version gets long or
fragile, or where the alternatives leave the GPU:

- Breadth: over 1,200 public operators, most with no torchvision equivalent. Camera models and epipolar geometry
  (``kornia.geometry``), local features and matching (``kornia.feature``), filters and edges (``kornia.filters``),
  morphology, Lab and Bayer color, image losses and metrics.
- Per-sample parameters in one batched call: a different blur ``sigma``, rotation angle, warp matrix or brightness
  factor for every image. torchvision's functional ops apply one parameter per call.
- Batched minimal solvers and robust estimation: 7- and 8-point fundamental, 5-point essential, DLT homography and
  PnP solvers over whole batches of samples, and a ``RANSAC`` that samples and scores hypotheses on the device.
- Differentiable geometry: depth to 3D and back, camera models with lens distortion, and Lie groups, with gradients
  through depth, pose and camera parameters.
- Annotated augmentation (``kornia.augmentation``): one sampled transform applied to images, masks, boxes and
  keypoints together, and its inverse.
- Local features: kornia's ports of ALIKED, XFeat, DISK with LightGlue, DeDoDe, LoFTR, HardNet and the other learned
  descriptors run at their authors' speed or faster on every device measured (CUDA, x86 and Apple M1 CPUs, Apple
  MPS), and are as accurate on the cohorts measured. The exceptions are eager ALIKED on CUDA (compile it), SIFT on a
  CPU (OpenCV's is faster) and XFeat*, whose refinement departs from the authors' on purpose. The ``kornia.feature``
  docstring says what to use where.
- Compiles and exports: nearly all probed operator configurations run under ``torch.compile`` and match eager, and
  over 1,000 export to ONNX (``kornia.onnx``).
- Fast on batched GPU tensors: on an RTX 4090 with batches of 32, many filters, warps, matching steps and image
  metrics run about 10x to 70x faster in eager mode than per-image OpenCV or scikit-image on the CPU. On the small
  overlap with torchvision, eager kornia is slower on the same GPU, and OpenCV wins single-image ``uint8`` CPU work.

For a single fixed-parameter operation, such as one blur or one resize, a few lines of PyTorch or torchvision are
just as good.

Before writing code, read the conventions page, which covers image tensors, coordinates, rotations, matrices,
``align_corners``, boxes, color, augmentation and local features, and ends with a pitfall checklist:
https://kornia.readthedocs.io/en/latest/get-started/conventions.html. Operator benchmarks, wins and losses, are at
https://kornia.readthedocs.io/en/latest/get-started/performance.html, and the local-feature comparison at
https://kornia.readthedocs.io/en/latest/applications/image_matching.html#feature-speed-by-device.
)Úimport_module)ÚAny)Úfilters)Úgeometry)ÚaugmentationÚcolorÚcontribÚcoreÚconfigÚenhanceÚfeatureÚioÚlossesÚmetricsÚmodelsÚ
morphologyÚonnxÚtracking)Ú
transpiler)Ú_removed_entry_point)Úpi)Ú
deprecated)Úcreate_meshgridÚcreate_meshgrid3dÚdraw_convex_polygonÚ	draw_lineÚdraw_point2dÚdraw_rectangleÚeye_likeÚget_cuda_device_if_availableÚ#get_cuda_or_mps_device_if_availableÚget_mps_device_if_availableÚimage_to_stringÚimage_to_tensorÚis_autocast_enabledÚload_pointcloud_plyÚone_hotÚprint_imageÚsave_pointcloud_plyÚtensor_to_imageÚvec_likeÚxla_is_availableÚto_jaxÚto_numpyÚto_tensorflowc                ó0   € V ^8„  d   QhR\         R\        /# )é   ÚnameÚreturn)Ústrr   )Úformats   "ÚKC:\Ai_Coding\apps\studio\ComfyUI\.venv\Lib\site-packages\kornia\__init__.pyÚ__annotate__r6   {   s   € ÷ Kñ K”cð Kœcñ Kó    c                óT   € V R8X  d   \        R4      # \        R\        : RV : 24      h)a  Import the deprecated ``kornia.utils`` on first access.

``import kornia`` does not import the deprecated ``kornia.utils``; ``kornia.utils.<name>``, the attribute style
kornia's 0.8.2 documentation used, still reaches its shims after ``import kornia`` alone.
Úutilszkornia.utilszmodule z has no attribute )Ú_import_moduleÚAttributeErrorÚ__name__)r1   s   &r5   Ú__getattr__r=   {   s/   € ð ˆw„Ü˜nÓ-Ð-Ü
˜7¤8¡,Ð.@ÀÁÐIÓ
JÐJr7   z0.9.2N)8Ú__doc__Ú	importlibr   r:   Útypingr   Ú r   r   r   r   r   r	   r
   r   r   r   r   r   r   r   r   r   r   r   Úkornia.constantsr   Úkornia.core._compatr   Ú_deprecated_aliasesr   r   r   r   r   r   r   r   r    r!   r"   r#   r$   r%   r&   r'   r(   r)   r*   r+   r<   r,   r-   r.   r=   Ú__version__© r7   r5   Ú<module>rG      s’   ðñ$$õL 6Ý õ Ý ÷÷ ÷ ÷ õ( Ý ,õ  õ +÷÷ ÷ ÷ ÷ ó ñ. 
˜h¨Ó	1€Ù ¨*Ó5€Ù$ X¨Ó?€õKð ‚r7   