3jI%SSKJr SSKrSSKJr 0r\\\4\S'S\S'S\S'S \S '"S S \R5r"S S\R5r "SS\R5r g))DictN)nnurlszXhttps://github.com/ducha-aiki/Key.Net-Pytorch/raw/main/model/HyNet/weights/HyNet_LIB.pthlibertyzWhttps://github.com/ducha-aiki/Key.Net-Pytorch/raw/main/model/HyNet/weights/HyNet_ND.pth notredamezXhttps://github.com/ducha-aiki/Key.Net-Pytorch/raw/main/model/HyNet/weights/HyNet_YOS.pthyosemitec ^\rSrSrSrSS\S\S\S\S\SS 4 U4S jjjrSS jr S\ 4S jr S \ RS\ R4SjrSrU=r$)FilterResponseNorm2daxFeature Response Normalization layer from 'Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks', see :cite:`FRN2019` for more details. .. math:: y = \gamma \times \frac{x}{\sqrt{\mathrm{E}[x^2]} + |\epsilon|} + \beta Args: num_features: number of channels eps: normalization constant is_bias: use bias is_scale: use scale drop_rate: dropout rate, is_eps_leanable: if eps is learnable Returns: torch.Tensor: Normalized features Shape: - Input: :math:`(B, \text{num_features}, H, W)` - Output: :math:`(B, \text{num_features}, H, W)` num_featuresepsis_biasis_scaleis_eps_leanablereturnNc>[TU]5 XlX lXPlX0lX@l[R"[R"SUSS5SS9Ul [R"[R"SUSS5SS9Ul U(a/[R"[R"S5SS9UlO'UR!S[R"U/55 UR#5 g)NT requires_gradr )super__init__r init_epsrrrr Parametertorchonesweightzerosbiastensorr register_bufferreset_parameters)selfr r rrr __class__s N/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/feature/hynet.pyrFilterResponseNorm2d.__init__6s ( .  ll5::aq!#DTXY LLQ a!CSWX ||ELLO4HDH   cU(; < c4[RRUR5 [RR UR 5 UR (a5[RRURUR5 ggN) rinitones_rzeros_rr constant_r rr"s r$r!%FilterResponseNorm2d.reset_parametersNsT  dkk" tyy!    GG  dhh 6 r&c:SR"S0URD6$)Nz+num_features={num_features}, eps={init_eps}format__dict__r-s r$ extra_reprFilterResponseNorm2d.extra_reprTs<CCTdmmTTr&xc,URS5RSS/SS9nU[R"X RR 5-5-nUR (aURU-nUR(aXR-nU$)NT)dimkeepdim) powmeanrrsqrtr absrrrr)r"r6nu2s r$forwardFilterResponseNorm2d.forwardWsoeeAhmmAm5  C((,,.01 1 == aA <<II Ar&)rr rrrrr r)gư>TTFrN)__name__ __module__ __qualname____firstlineno____doc__intfloatboolrr!strr4rTensorrA__static_attributes__ __classcell__r#s@r$r r s6 %             07 UCU  %,,  r&r c^\rSrSrSrS\SS4U4SjjrS SjrS\4Sjr S \ RS\ R4S jr S r U=r$) TLUfaTLU layer from 'Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks, see :cite:`FRN2019` for more details. :math:`{\tau}` is learnable per channel. .. math:: y = \max(x, {\tau}) Args: num_features: number of channels Returns: torch.Tensor Shape: - Input: :math:`(B, \text{num_features}, H, W)` - Output: :math:`(B, \text{num_features}, H, W)` r rNc>[TU]5 Xl[R"[ R "SUSS5*SS9UlUR5 g)z:max(y, tau) = max(y - tau, 0) + tau = ReLU(y - tau) + tau.rTrN) rrr rrrrtaur!)r"r r#s r$r TLU.__init__ysE (<<A|Q!B BRVW r&cX[RRURS5 g)N)rr)r,rUr-s r$r!TLU.reset_parameterss $((B'r&c:SR"S0URD6$)Nznum_features={num_features}r0r1r-s r$r4TLU.extra_reprs,33DdmmDDr&r6cB[R"XR5$r()rmaxrUr"r6s r$rA TLU.forwardsyyHH%%r&)r rUrC)rDrErFrGrHrIrr!rLr4rrMrArNrOrPs@r$rRrRfsN$ S T (ECE&&%,,&&r&rRc^\rSrSrSrSrSS\S\S\S\S\S \S S 4U4S jjjr S \ RS \ R4Sjr Sr U=r$)HyNetann.Module, which computes HyNet descriptors of given grayscale patches of 32x32. This is based on the original code from paper "HyNet: Learning Local Descriptor with Hybrid Similarity Measure and Triplet Loss". See :cite:`hynet2020` for more details. Args: pretrained: Download and set pretrained weights to the model. is_bias: use bias in TLU layers is_bias_FRN: use bias in FRN layers dim_desc: descriptor dimensionality, drop_rate: dropout rate, eps_l2_norm: to avoid div by zero Returns: HyNet descriptor of the patches. Shape: - Input: :math:`(B, 1, 32, 32)` - Output: :math:`(B, 128)` Examples: >>> input = torch.rand(16, 1, 32, 32) >>> hynet = HyNet() >>> descs = hynet(input) # 16x128 pretrainedr is_bias_FRNdim_desc drop_rate eps_l2_normrNc >[TU]5 X`lX@lXPl[ R "[SUS9[S5[ R"SSSSUS9[SUS9[S55Ul [ R "[ R"SSSSUS9[SUS9[S55Ul [ R "[ R"SSSSSUS9[SUS9[S55Ul [ R "[ R"SSSSUS9[SUS9[S55Ul [ R "[ R"SS SSSUS9[S US9[S 55Ul[ R "[ R"S S SSUS9[S US9[S 55Ul[ R "[ R "UR5[ R"S URS S S 9[ R""URS S 95Ul[ R&"SUR-SUR-SS5UlU(aI[*R,R/[0S[*R2"S5S9nUR5USS9 UR75 g)Nr)rrcr9) kernel_sizepaddingr@r8)rjstriderkrF)rjr)affineg@g?grcpu) map_locationT)strict)rrrhrfrgr Sequentialr rRConv2dlayer1layer2layer3layer4layer5layer6Dropout BatchNorm2dlayer7LocalResponseNorm desc_normrhubload_state_dict_from_urlrdeviceload_state_dicteval) r"rdrrerfrgrhpretrained_dictr#s r$rHyNet.__init__s> & "mm K 8 F IIaAG D [ 9 G   mm IIb"!QW E [ 9 G  mm IIb"!Aqw O [ 9 G  mm IIb"!QW E [ 9 G  mm IIb#1Q P k : H  mm IIc3Aqw G k : H  mm JJt~~ & IIc4==ae D NN4== 7  --a$--.?t}}ATVY[^_ #ii@@i_d_k_klq_r@sO   > r&r6cpURU5nURU5nURU5nURU5nUR U5nUR U5nUR U5nURXR-5nURURS5S5nU$)NrrX) rvrwrxryrzr{r~rrhviewsizer^s r$rA HyNet.forwards KKN KKN KKN KKN KKN KKN KKN NN1/// 0 FF166!9b !r&) rrfrgrhrvrwrxryrzr{r~)FTTrng333333?g|=)rDrErFrGrH patch_sizerKrIrJrrrMrArNrOrPs@r$raras8J! "=== =  =  == ==~  %,,  r&ra) typingrrrrrL__annotations__Moduler rRrar0r&r$rsu$ d38nlYm[mZF299FR"&"))"&JhBIIhr&