3jj%SSKJrJr SSKrSSKJs Jr SSKJr SSKJ r SSK J r SSK J r SSKJrJrJr "S S \ 5rS S S \"5S.r\\S'Sr"SS\R,5r"SS\R,5r"SS\R25rS#S\S\S\S\S\S\S\R24Sjjr"SS \R,5r"S!S"\5rg)$)ListOptionalN)nn) TypedDict)SpatialGradient)pyrdown)Detector_configMultiResolutionDetectorget_default_detector_configcB\rSrSr%Sr\\S'\\S'\\S'\\S'Srg) KeyNet_confzDefine the configuration schema for the KeyNet feature detector. Attributes: num_filters: The number of filters in the convolutional layers. num_levels: The number of levels in the image pyramid. num_filters num_levels kernel_size Detector_confN) __name__ __module__ __qualname____firstlineno____doc__int__annotations__r __static_attributes__rO/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/feature/keynet.pyrrs O""rr)rrrrkeynet_default_configzXhttps://github.com/axelBarroso/Key.Net-Pytorch/raw/main/model/weights/keynet_pytorch.pthcj^\rSrSrSrSU4SjjrS\RS\R4SjrSr U=r $) _FeatureExtractor9zNHelper class for KeyNet. It loads both, the handcrafted and learnable blocks returnc^>[TU]5 [5Ul[ 5UlgN)super__init___HandcraftedBlockhc_block_LearnableBlocklb_blockself __class__s rr*_FeatureExtractor.__init__?s" )+ ') rxcJURU5nURU5nU$r(r,r.)r0r3x_hcx_lbs rforward_FeatureExtractor.forwardEs#}}Q}}T" rr5r&N rrrrrr*torchTensorr8r __classcell__r1s@rr$r$9s- * %,,rr$cj^\rSrSrSrSU4SjjrS\RS\R4SjrSr U=r $) r+KzaHelper class for KeyNet, it defines the handcrafted filters within the Key.Net handcrafted block.r&cD>[TU]5 [SS5Ulg)Nsobelr )r)r*rspatial_gradientr/s rr*_HandcraftedBlock.__init__Ns  / ;rr3c URU5nUSS2SS2SSS2SS24USS2SS2SSS2SS24pCURU5nUSS2SS2SSS2SS24USS2SS2SSS2SS24pvURU5SS2SS2SSS2SS24n[R"X4UR5UR5X4-XwR5XhXh-/ S5n U $)Nrr )rDr<catsquare) r0r3rCdxdysobel_dxdxxdxydyyhc_featss rr8_HandcraftedBlock.forwardRs%%a(q!Q1}%uQ1a]';B((,Aq!QM*HQ1a],CS##B'1aA 699bbiik299;jjl\_fifoprstr)rDr:r;r?s@rr+r+Ks+k<  %,,  rr+cv^\rSrSrSrS S\SS4U4SjjjrS\RS\R4Sjr S r U=r $) r-azMHelper class for KeyNet. It defines the learnable blocks within the Key.Net in_channelsr&Nc~>[TU]5 [U5Ul[5Ul[5Ulgr()r)r*_KeyNetConvBlockconv0conv1conv2)r0rSr1s rr*_LearnableBlock.__init__gs. %k2 %' %' rr3cdURURURU555nU$r()rXrWrV)r0r3s rr8_LearnableBlock.forwardns& JJtzz$**Q-0 1r)rVrWrX) ) rrrrrrr*r<r=r8rr>r?s@rr-r-as@ (C(((%,,rr-rS out_channelsrstridepaddingdilationr&c [R"[R"XX#XE5[R"U5[R"SS95$)zMCreate KeyNet Conv Block. Default learnable convolutional block for KeyNet. Tinplace)r SequentialConv2d BatchNorm2dReLU)rSr]rr^r_r`s rrUrUss= == +['T |$  rc^\rSrSrSrS S\S\\SS4U4SjjjrS\ RS\ R4S jr S r U=r $) KeyNeta0Key.Net model definition -- local feature detector (response function). This is based on the original code from paper "Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters". See :cite:`KeyNet2019` for more details. .. image:: _static/img/KeyNet.png Args: pretrained: Download and set pretrained weights to the model. keynet_conf: Dict with initialization parameters. Do not pass it, unless you know what you are doing`. Returns: KeyNet response score. Shape: - Input: :math:`(B, 1, H, W)` - Output: :math:`(B, 1, H, W)` N pretrained keynet_confr&c >[TU]5 Uc[nUSnUSUlUSnUS-n[ 5Ul[ R"[ R"X0R-SXES9[ R"SS95Ul U(aI[RR[[R"S 5S 9nUR!US SS 9 UR#5 g) Nrrrr )rSr]rr_Trbcpu) map_location state_dict)strict)r)r*r"rr$feature_extractorrrdrerg last_convr<hubload_state_dict_from_url KeyNet_URLdeviceload_state_dicteval)r0rkrlrrr_pretrained_dictr1s rr*KeyNet.__init__s   /K!-0 %l3!-0 "!2!4 II'//9Wb  GGD !   #ii@@Z_ZfZfglZm@nO  !>t L rr3crURSURSp2URU5/n[SUR5HGn[ USS9nURU5n[ R "XbU4SSS9nURU5 MI UR[R"US55nU$) zX - input image.rnr r g333333?)factorbilinearF)sizemode align_corners) shapersrangerrF interpolateappendrtr<rG)r0r3hwfeats_feats_iscoress rr8KeyNet.forwardswwqz1771:1%)%;%;A%>$?q$//*A#&A,,Q/GmmGa&zY^_G LL ! +  % 34 r)rsrtr)FN)rrrrrboolrrr*r<r=r8rr>r?s@rririsM*4h{>S_c.  %,,  rric^\rSrSrSrSS\S\S\\S\\ RS\\ RS \S \ S S4U4S jjjr S r U=r$)KeyNetDetectoraNMulti-scale feature detector based on KeyNet. This is based on the original code from paper "Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters". See :cite:`KeyNet2019` for more details. .. image:: _static/img/keynet.jpg Args: pretrained: Download and set pretrained weights to the model. num_features: Number of features to detect. keynet_conf: Dict with initialization parameters. Do not pass it, unless you know what you are doing`. ori_module: for local feature orientation estimation. Default: :class:`~kornia.feature.PassLAF`, which does nothing. See :class:`~kornia.feature.LAFOrienter` for details. aff_module: for local feature affine shape estimation. Default: :class:`~kornia.feature.PassLAF`, which does nothing. See :class:`~kornia.feature.LAFAffineShapeEstimator` for details. Nrk num_featuresrl ori_module aff_module compile_modelscore_thresholdr&c X>Uc[n[X5n[T U] XUSXEXg5 g)Nr)r"rir)r*) r0rkrrlrrrrmodelr1s rr*KeyNetDetector.__init__s6  /Kz/  _!=zWd rr)FiNNNFg)rrrrrrrrrrModulefloatr*rr>r?s@rrrs*! -1*.*.#!$   k*  RYY'  RYY'       rr)rrr!r rnr )typingrrr<torch.nn.functionalr functionalrtyping_extensionsrkornia.filtersrkornia.geometry.transformrscale_space_detectorr r r rr"rrwrr$r+rdr-rrUrirrrrrs $"! '*-gg #) # 02 &{h  $ ,bmm&       ]]&7RYY7t# ,# r