3j#SSKJrJrJr SSKrSSKJr SSKJr SSKJ r J r SSK J r SSK Jr SSKJr "S S \R"5rg) )DictOptionalTupleN)nn)rgb_to_grayscale)LocalFeatureMatcherLoFTR)find_homography_dlt_iterated)RANSAC)warp_perspectivec P^\rSrSrSrS"S\R S\S\SS4U4SjjjrS \ RS \ RS\ R4S jr S \ RS \ RS\ R4Sjr S\ RS\ RS\ RS\ R4Sjr S\ RS\ RS\\\ R44SjrS\ RS\ RS\ R4SjrS\\\ R4S\\\ R44SjrS#S\ RS\ RS\\ RS\\ RS\\ R\ R44 SjjrS\ RS\ R4S jrS!rU=r$)$ ImageStitcheraStitch two images with overlapping fields of view. Args: matcher: image feature matching module. estimator: method to compute homography, either "vanilla" or "ransac". "ransac" is slower with a better accuracy. blending_method: method to blend two images together. Only "naive" is currently supported. Note: Current implementation requires strict image ordering from left to right. .. code-block:: python IS = ImageStitcher(KF.LoFTR(pretrained='outdoor'), estimator='ransac').cuda() # Compute the stitched result with less GPU memory cost. with torch.inference_mode(): out = IS(img_left, img_right) # Show the result plt.imshow(K.tensor_to_image(out)) matcher estimatorblending_methodreturnNc>[TU]5 XlX lX0lUS;a[ SUS35eUS:Xa[ S5Ulgg)N)ransacvanillaUnsupported estimator $. Use `ransac` or `vanilla` instead.r homography)super__init__rrrNotImplementedErrorr r)selfrrr __class__s X/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/contrib/image_stitching.pyrImageStitcher.__init__6sV  ". 1 1%(>ykIm&no o   .DK ! keypoints1 keypoints2c  URS:Xa1[USUS[R"USSS2S455nU$URS:XaUR X!5up4USnU$[ SURS35e)zEstimate homography by the matched keypoints. Args: keypoints1: matched keypoint set from an image, shaped as :math:`(N, 2)`. keypoints2: matched keypoint set from the other image, shaped as :math:`(N, 2)`. rNrrrr)rr torch ones_likerr)rr"r#homo_s r_estimate_homography"ImageStitcher._estimate_homography@s >>Y &/4 *T"2EOOJtUVXYzDZ4[D ^^x 'kk*9GD:D &(>t~~>NNr&st tr!argskwargsc USUSUSpTn[[UR555Vs/sHo`RX5U:HXEU:H5PM nn[U5S:Xa [ S5e[ R "U5$s snf)z%Compute the corresponding homography. keypoints0r" batch_indexesrz6Compute homography failed. No matched keypoints found.)rangelenuniquer) RuntimeErrorr%cat)rr+r,kp1kp2idxihomoss restimate_transform ImageStitcher.estimate_transformSs|,f\.BF?D[#RWX[\_\f\f\hXiRjkRjQ**3ax=#Qh-HRjk u:?WX Xyy ls%B src_imgdst_imgmaskcURS:Xa[R"US:HX5nU$[SURS35e)zBlend two images together.naivezUnsupported blending method z. Use `naive`.)rr%wherer)rr<r=r>outs r blend_imageImageStitcher.blend_image]sL   7 *++dai:C &(DTEYEYDZZh&ij jr!image_1image_2c[UR[[45(a[ U5[ U5S.nU$[ SURS35e)z(Preprocess input to the required format.)image0image1zThe preprocessor for z has not been implemented.) isinstancerr rrr)rrFrG input_dicts r preprocessImageStitcher.preprocessfsV dllU,?$@ A A*73*73J &(=dll^Ke&fg gr!imagecURS5n[UR5RS5R 5R 5R 55nUS:XaU$USSU24$)a4Crop redundant empty columns from a stitched panorama. Args: image: Stitched image tensor, typically shaped :math:`(B, C, H, W_{panorama})`, where ``W_{panorama}`` is the panorama width after concatenation/warping. mask: Validity mask aligned with ``image`` where non-zero values mark pixels covered by at least one input view. Returns: The stitched image cropped to remove trailing invalid columns (columns that contain no valid pixels in the mask). If no redundant area is found, the original ``image`` is returned. )rrAr.N)sumintboolanylongargminitem)rrOr>mask_indexs r postprocessImageStitcher.postprocessrsg  EJJL$$Q',,.557<<>? A:LS&5&[!!r!datac$URU5$)aRRun the configured feature matcher on preprocessed inputs. Args: data: Matcher input dictionary, usually containing ``image0`` and ``image1`` tensors. Returns: A correspondence dictionary produced by ``self.matcher``. Typical entries are matched keypoints and batch indices used later for homography estimation. )r)rr\s r on_matcherImageStitcher.on_matchers||D!!r! images_left images_right mask_left mask_rightcURX5nURSURSURS-4nURU5nUR"S0UD6n[ X(U5n [ R "U[ R"U5/S5n Uc[ R"U5nUc[ R"U5n[ XHUSS9n [ R "U[ R"U5/S5n URXU 5X-R5RU R54$)aStitch two images into a shared panorama frame. Args: images_left: Reference image tensor, shaped :math:`(B, C, H, W_l)`. images_right: Image tensor to be warped onto the reference frame, shaped :math:`(B, C, H, W_r)`. mask_left: Optional validity mask for ``images_left`` with the same shape as ``images_left``. mask_right: Optional validity mask for ``images_right`` with the same shape as ``images_right``. Returns: A tuple ``(stitched_image, stitched_mask)``: - ``stitched_image`` is the blended panorama tensor with width ``W_l + W_r`` before final cropping. - ``stitched_mask`` is a mask tensor indicating which panorama pixels are valid after warping and blending. nearest)mode) rMshaper^r:r r%r4 zeros_liker&rDrStodtype) rr`rarbrcrL out_shapecorrespondencesr'r<r=src_maskdst_masks r stitch_pairImageStitcher.stitch_pairs4__[?  &&r*K,=,=b,ALDVDVWYDZ,Z[ //*5&&99"Q=W=W=Y=\=\]e]k]k=lllr!imgscUSn[R"U5n[[U5S- 5HnUR X!US-U5up#M UR X#5$)aIteratively stitch a sequence of overlapping images. Args: *imgs: Ordered image tensors from left to right. Each tensor is expected to share a common height and overlap with the next image. Each tensor typically follows :math:`(B, C, H, W)`. Returns: A single panorama tensor obtained by stitching all provided images in order and cropping trailing empty space. rrA)r%r&r0r1rrrZ)rrtimg_outrbr8s rforwardImageStitcher.forwardsbq'OOG, s4y1}%A!%!1!1'A; !R GY&33r!)rrrr)rr@)NN)__name__ __module__ __qualname____firstlineno____doc__rModulestrrr%Tensorr)r:rDrrMrZr^rrrrrw__static_attributes__ __classcell__)rs@rrrs./ /c/WZ/im//u||Z_ZfZf&     QVQ]Q] 5<<%,,ell_d_k_k %,,  $sTYT`T`O`Ja ""U\\"ell", "tC$56 "4U\\@Q;R "$-1-1 )m\\)mll)mELL) )m U\\* )m u||U\\) * )mV4U\\4ell44r!r)typingrrrr%r kornia.colorrkornia.featurerr kornia.geometry.homographyr kornia.geometry.ransacr kornia.geometry.transformr r~rrir!rrs4$)( )5C)6s4BIIs4r!