3jG`rSSKJrJrJrJrJrJr SSKrSSKJ s J r SSKJ r SSK J r Jr SSKJr SSKJr SSKJr SS KJrJr S \R0S \R0S \R04S jrS \\\44SjrS%S\R0S\R0S\\R0S \R04SjjrS\R0S \\R0\R044SjrS%S\R0S\R0S\\R0S \\R0\R044SjjrS%S\R0S\R0S\\R0S \\R0\R044SjjrS&S\R0S\R0S\ S\\R0S \\R0\R044 Sjjr!S'S\R0S\R0S\ S\\R0S \\R0\R044 Sjjr"S(S\R0S\R0S\R0S\R0S\ S\ S\#S\\R0S \\R0\R044Sjjr$"SS \ RJ5r&"S!S"\ RJ5r'"S#S$\ RJ5r(g)))AnyClassVarDictListOptionalTupleN)nn)KORNIA_CHECK_DM_DESCKORNIA_CHECK_SHAPE)is_mps_tensor_safe)get_laf_center)DiscreteSteerer)get_adalam_default_config match_adalamd1d2returncH[R[R4n[U5(dF[U5(d6URU;a&URU;a[R "X5$US-R SSS9nUS-R SSS9nURSURS55URSURS55R5-SU-UR5-- nURSS9R5nU$) aaCompute pairwise L2 distances between rows of d1 and d2. Uses ``torch.cdist`` for float32/float64 on non-MPS devices. Falls back to a manual expand-and-norm implementation for MPS tensors and for half-precision dtypes (float16/bfloat16), since ``torch.cdist`` does not support half precision on CUDA and may be unavailable for these dtypes elsewhere. rT)dimkeepdimrg@g)min) torchfloat16bfloat16r dtypecdistsumrepeatsizetclampsqrt)rrhalfd1_sqd2_sqdms Q/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/feature/matching.py_cdistr* s MM5>> *D r " "-?-C-CY]I]bdbjbjrvbv{{2"" UKKAtK ,E UKKAtK ,E a $u||Arwwqz'B'D'D'F FrTVTXTXTZIZ ZB c    !B IcSSSS.nU$)Ng333333?F$@)thmutual spatial_thconfigs r)_get_default_fginn_paramsr42sE >F Mr+desc1desc2dm_cBUc [X5nU$[XU5 UnU$)aoCheck validity of provided distance matrix, or calculates L2-distance matrix if dm is not provided. Args: desc1: Batch of descriptors of a shape :math:`(B1, D)`. desc2: Batch of descriptors of a shape :math:`(B2, D)`. dm_: torch.Tensor containing the distances from each descriptor in desc1 to each descriptor in desc2, shape of :math:`(B1, B2)`. )r*r )r5r6r7r(s r)_get_lazy_distance_matrixr97s0 { E ! I U3/  Ir+r(c[R"SSURURS9n[R"SSUR[RS9nX4$)zOutput empty tensors. Returns: - Descriptor distance of matching descriptors, shape of :math:`(0, 1)`. - Long torch.Tensor indexes of matching descriptors in desc1 and desc2, shape of :math:`(0, 2)`. rr)devicerr)remptyr;rlong)r(distsidxss r) _no_matchr@KsD KK1RYYbhh ?E ;;q!BIIUZZ @D ;r+c[USS/5 [USS/5 [U5S:Xd[U5S:Xa [U5$[XU5n[R "USS9upE[R "SURS5URS9n[R"URSS5URSS5/S5nURSS5URSS54$) aMFind nearest neighbors in desc2 for each vector in desc1. If the distance matrix dm is not provided, :py:func:`torch.cdist` is used. Args: desc1: Batch of descriptors of a shape :math:`(B1, D)`. desc2: Batch of descriptors of a shape :math:`(B2, D)`. dm: torch.Tensor containing the distances from each descriptor in desc1 to each descriptor in desc2, shape of :math:`(B1, B2)`. Returns: - Descriptor distance of matching descriptors, shape of :math:`(B1, 1)`. - Long torch.Tensor indexes of matching descriptors in desc1 and desc2, shape of :math:`(B1, 2)`. BDIMrrrr;r) r lenr@r9rraranger!r;catview)r5r6r(distance_matrix match_dists idxs_in_2idxs_in1 matches_idxss r)match_nnrPXs$usEl+usEl+ E aSZ1_/bAO"YYA>K||Ay~~a09I9IJH99hmmB2INN2q4IJANL   B "L$5$5b!$< <B"YYA>K#iiQ?L<<+A+ABLA/"6"6q"99!y%9#2%>> yy,"3"3B":INN2q> yy)..Q"79J9J2q9Q!RTUVWab ".   B "L$5$5b!$< <rFN)r r[r@r9r`r"rGflipr rrfloattorepeat_interleaver!r absrreshaperrJsortmax)r5r6r.r(rKdists1idx1dists2idx2idxs_dm idxs1_rep mutual_idxs1 mutual_idxs2 good_idxs1 good_idxs2 dists1_good dists2_good_idx_upl1idx_upl2rLrOs r) match_smnnrzsv,usEl+usEl+ A A 2/bAOU2?LFU2/@/@/BCLF 6{Q3v;?yy|!$''kk$**, DG88118MIwwu~,,TYYq\1= INNPTTYZT[GoodiilDIIaLAG{{q{)!,t3 {{q{)!,t3 ++B/0 ++B/0 ..r23 ..r23 jjAqD!12 8jjAqD!12 8) ii H 5{87LM ! $/$4$4R$;\=N=NrST=U\   $$%.o$>! \  $$r+lafs1lafs2r0r/c:[USS/5 [USS/5 Sn[XU5n U Rn URSS:a [ U 5$[ S[ SURS55n [R"XSSS 9upU S S 2S4n[U5RS S5nXn[R"UUSS- SSS 9nU S S 2SS 24US S 2SS 24U:RU 5U--nUR SS 9unnUnU S S 2S4n UU- nUU:*nUUn[U5S:Xa [ U 5$[R"SU RS5U R S9UnU Un [R""URS S5U RS S5/S5nURS S5URS S5nnU(dUU4$[R "U SS 9unnUS S 2S4UUS S 2S4:HnUUUU4$)a&Find nearest neighbors in desc2 for each vector in desc1. The method satisfies first to second nearest neighbor distance <= th, and assures 2nd nearest neighbor is geometrically inconsistent with the 1st one (see :cite:`MODS2015` for more details) If the distance matrix dm is not provided, :py:func:`torch.cdist` is used. Args: desc1: Batch of descriptors of a shape :math:`(B1, D)`. desc2: Batch of descriptors of a shape :math:`(B2, D)`. lafs1: LAFs of a shape :math:`(1, B1, 2, 3)`. lafs2: LAFs of a shape :math:`(1, B2, 2, 3)`. th: distance ratio threshold. spatial_th: minimal distance in pixels to 2nd nearest neighbor. mutual: also perform mutual nearest neighbor check. dm: torch.Tensor containing the distances from each descriptor in desc1 to each descriptor in desc2, shape of :math:`(B1, B2)`. Return: - Descriptor distance of matching descriptors, shape of :math:`(B3, 1)`. - Long torch.Tensor indexes of matching descriptors in desc1 and desc2. Shape: :math:`(B3, 2)`, where 0 <= B3 <= B1. rBrCg.Arr rFrYNrF)rbrrDrE)r r9rr[r@rjrrr\r rJnormrerGrHr!r;rI)r5r6r{r|r.r0r/r( BIG_NUMBERrKrnum_candidates vals_candrMr]xy2 candidates_xykdist fginn_valsfginn_vals_best_fginn_idxs_bestvals_2ndr^r_rLrNrOrw idxs_in_1_mut good_masks r) match_fginnrs:FusEl+usEl+J/bAO  ! !E {{1~))CEKKN34N ::o1V[\I QT?D   $ $R +CNM JJ}}Qq'99QA FE1ab5!U1ab5\J%>$B$B5$IJ$VVJ(21(=%O% H!Q$I 8OE B;D+K ;1))||Ay~~a09O9OPQUVH$I99hmmB2INN2q4IJANL + 0 0Q 79J9J2q9QK L((yya8A}QT"mLA4F&GGI y !< #: ::r+c^\rSrSrSrS S\S\SS4U4SjjjrS\RS \RS\ \R\R44S jr S r U=r $) DescriptorMatcheriKa nn.Module version of descriptor-only matching functions. This matcher only requires descriptors (no LAFs). For geometry-aware matching that uses LAFs, see :class:`~kornia.feature.GeometryAwareDescriptorMatcher`. See :func:`~kornia.feature.match_nn`, :func:`~kornia.feature.match_snn`, :func:`~kornia.feature.match_mnn` or :func:`~kornia.feature.match_smnn` for more details. Args: match_mode: type of matching, can be `nn`, `snn`, `mnn`, `smnn`. th: threshold on distance ratio, or other quality measure. match_moder.rNc>[TU]5 UR5n/SQUlX0R;a[ USUR35eX0lX lg)Nr mnnsnnsmnn is not supported. Try one of )super__init__lower known_modesNotImplementedErrorrr.)selfrr. _match_mode __class__s r)rDescriptorMatcher.__init__ZsY %++- 7 .. .% 4RSWScScRd&ef f%r+r5r6c"URS:Xa [X5nU$URS:Xa [X5nU$URS:Xa[XUR5nU$URS:Xa[ XUR5nU$[ e)anRun forward. Args: desc1: Batch of descriptors of a shape :math:`(B1, D)`. desc2: Batch of descriptors of a shape :math:`(B2, D)`. Returns: - Descriptor distance of matching descriptors, shape of :math:`(B3, 1)`. - Long torch.Tensor indexes of matching descriptors in desc1 and desc2, shape of :math:`(B3, 2)` where :math:`0 <= B3 <= B1`. r rrrrrPrWr`r.rzr)rr5r6outs r)forwardDescriptorMatcher.forwardcs ??d "5(C __ %E)C __ %E$''2C  __ &U4773C & %r+)rrr.)r皙?)__name__ __module__ __qualname____firstlineno____doc__strrdrrTensorrr__static_attributes__ __classcell__rs@r)rrKsc 3EDU\\%,,5W\WcWcIcCdr+rct^\rSrSrSrSS\S\S\S\S\SS 4 U4S jjjr SS \ RS \ RS \ \ RS\ \ R\ R44SjjrSS\ RS\ RS\S\ \S\ \ R\ R\ \44 SjjrSrU=r$)DescriptorMatcherWithSteereri}a_Matching that is invariant under rotations, using Steerers. Args: steerer: An instance of :func:`kornia.feature.steerers.DiscreteSteerer`. steerer_order: order of discretisation of rotation angles, e.g. 4 leads to quarter rotations. steer_mode: can be `global`, `local`. `global` means that the we output matches from the global rotation with most matches. `local` means that we output matches from a distance matrix where the distance between each descriptor pair is the minimal over rotations. match_mode: type of matching, can be `nn`, `snn`, `mnn`, `smnn`. WARNING: using steer_mode `global` with match_mode `nn` will lead to bad results since `nn` doesn't generate different amount of matches depending on goodness of fit. th: threshold on distance ratio, or other quality measure. Example: >>> import kornia as K >>> import kornia.feature as KF >>> device = K.core.utils.get_cuda_or_mps_device_if_available() >>> img1 = torch.randn([1, 3, 768, 768], device=device) >>> img2 = torch.randn([1, 3, 768, 768], device=device) >>> dedode = KF.DeDoDe.from_pretrained(detector_weights="L-C4-v2", descriptor_weights="B-SO2").to(device) >>> steerer_order = 8 # discretisation order of rotation angles >>> steerer = KF.steerers.DiscreteSteerer.create_dedode_default( ... generator_type="SO2", steerer_order=steerer_order ... ) >>> steerer = steerer.to(device) >>> matcher = KF.matching.DescriptorMatcherWithSteerer( ... steerer=steerer, steerer_order=steerer_order, steer_mode="global", match_mode="smnn", th=0.98 ... ) >>> with torch.inference_mode(): ... kps1, scores1, descs1 = dedode(img1, n=20_000) ... kps2, scores2, descs2 = dedode(img2, n=20_000) ... kps1, kps2, descs1, descs2 = kps1[0], kps2[0], descs1[0], descs2[0] ... dists, idxs, num_rot = matcher( ... descs1, descs2, normalize=True, subset_size=1000, ... ) >>> # print(f"{idxs.shape[0]} tentative matches with steered DeDoDe") >>> # print(f"at rotation of {num_rot * 360 / steerer_order} degrees") steerer steerer_order steer_moderr.rNcf>[TU]5 XlX lUR 5nSS/UlX`R ;a[ USUR 35eX`lUR 5n/SQUlXpR;a[ USUR35eXpl XPl g)Ngloballocalrr) rrrrrknown_steer_modesrrrrr.) rrrrrr. _steer_moderrs r)r%DescriptorMatcherWithSteerer.__init__s  *%++- "*G!4 44 4% 4RSWSiSiRj&kl l%%++- 7 .. .% 4RSWScScRd&ef f%r+rrr(c URS:Xa [XUS9$URS:Xa [XUS9$URS:Xa[XURUS9$URS:Xa[ XURUS9$[ e)Nr )r(rrrr)rrrr(s r)matching_function.DescriptorMatcherWithSteerer.matching_functionsz ??d "Br* * __ %R+ + __ %RTWW4 4 __ &bdgg"5 5% %r+r5r6 normalize subset_sizecSnU(a*[R"USS9n[R"USS9nURS:XGaUb[R"UR S5SUn[R"UR S5SUnU"XX'US9u pUR RUUUS9nURXS5upXU4$URXS5upSn[SUR5HVn UR RXS9nURXS5upU R SU R S:dMRXU pZn MX OURS :Xaw[X5n[SUR5H=nUR RXS9n[X5n[R"X5nM? URXU5upO[eXU4$) a+Run forward. Args: desc1: Batch of descriptors of a shape :math:`(B1, D)`. desc2: Batch of descriptors of a shape :math:`(B2, D)`. normalize: bool to decide whether to F.normalize descriptors to unit norm. subset_size: If set, the subset size to use for determining optimal number of rotations. Smaller subset size leads to faster but less accurate matching. Only used when `self.steer_mode` is `"global"`. Returns: - Descriptor distance of matching descriptors, shape of :math:`(B3, 1)`. - Long torch.Tensor indexes of matching descriptors in desc1 and desc2, shape of :math:`(B3, 2)` where :math:`0 <= B3 <= B1`. - Number of global rotations from desc1 to desc2, in terms of `self.steerer_order` (will be `None` if `self.steer_mode` is `local`). NrFrDrr)r) steerer_powerrrr)Frrrrandpermr[rsteer_descriptionsrrangerr*minimumr)rr5r6rrrot1to2 subsample1 subsample2rwdistidxrdist_newidx_newr(dm_news r)r$DescriptorMatcherWithSteerer.forwards2 KK2.EKK2.E ??h &&"^^EKKN;L[I "^^EKKN;L[I $%%'! 1  77")'8 !225F '))..uTBIDG1d001 777S$($:$:5$N!==#ciil2)1AwDw 2 __ '%B1d001 777S-]]2.2..uR@ID#% %'!!r+)rrrrrrr.)rrrN)FN)rrrrrrintrrdrrrrrrboolrrrrs@r)rr}s'Z#        2PT &,, &$)LL &6>u||6L & u||U\\) * &$ %) @"||@"||@" @" c] @" u||U\\8C=8 9 @"@"r+rc $^\rSrSr%SrSS/r\\\\ S'SS\S\ \ \\ R4S S4U4S jjjrS \ RS \ RS \ RS\ RS \\ R\ R44 SjrSrU=r$)GeometryAwareDescriptorMatcheriann.Module version of geometry-aware matching functions that use LAFs (Local Affine Frames). Unlike :class:`~kornia.feature.DescriptorMatcher`, this matcher requires both descriptors and LAFs. See :func:`~kornia.feature.match_fginn` or :func:`~kornia.feature.match_adalam` for more details. Args: match_mode: type of matching, can be `fginn` or `adalam`. params: dictionary of parameters for the matching function. fginnadalamrNrparamsrc>[TU]5 UR5nX0R;a[ USUR35eX0lU=(d 0Ulg)Nr)rrrrrrr)rrrrrs r)r'GeometryAwareDescriptorMatcher.__init__sW %++- .. .% 4RSWScScRd&ef f%l r+r5r6r{r|c 0URS:Xa?[5nURUR5 [ XX4USUSUS5nU$URS:Xa2[ 5nURUR5 [ XX4US9nU$[e)aRun forward. Args: desc1: Batch of descriptors of a shape :math:`(B1, D)`. desc2: Batch of descriptors of a shape :math:`(B2, D)`. lafs1: LAFs of a shape :math:`(1, B1, 2, 3)`. lafs2: LAFs of a shape :math:`(1, B2, 2, 3)`. Returns: - Descriptor distance of matching descriptors, shape of :math:`(B3, 1)`. - Long torch.Tensor indexes of matching descriptors in desc1 and desc2, shape of :math:`(B3, 2)` where :math:`0 <= B3 <= B1`. rr.r0r/rr2)rr4updaterrrrr)rr5r6r{r|rr_paramss r)r&GeometryAwareDescriptorMatcher.forward&s" ??g %.0F MM$++ &eE&,|H\^dem^noC __ (/1G NN4;; 'uU'JC & %r+)rr)rN)rrrrrrrrr__annotations__rrrrrrrrrrs@r)rrs )0':K$s)$:#3#(4U\\HYCZ:[#gk##\\*/,,?D||TYT`T` u||U\\) *r+rr)rN)gffffff?N)rr-FN))typingrrrrrrrtorch.nn.functionalr functionalrkornia.core.checkr r kornia.core.utilsr kornia.feature.lafr kornia.feature.steerersrrrrrr*rr4r9r@rPrWrdr`rzrrModulerrrr1r+r)rs$>= F0-3;u||%,,$4S> MQ << % 3;ELL3I \\( %,, 5u||)C#D LP= <<= % =2:5<<2H= 5<< %&=<LP%= <<%= % %=2:5<<2H%= 5<< %&%=R]a%= <<%= % %=27%=CKELLCY%= 5<< %&%=R^b6% <<6% % 6%276%DLU\\DZ6% 5<< %&6%|!%J; <<J; <<J; <<J; << J;  J;  J; J; J; 5<< %&J;Z/ /dP"299P"f1RYY1r+