a db@sUddlmZmZddlZddlmZddlmmZddl m Z iZ ee e fe d<de d<de d<d e d <Gd d d ejZGd ddejZdS))CallableDictN)KORNIA_CHECK_SHAPEurlsz]https://github.com/DagnyT/hardnet/raw/master/pretrained/pretrained_all_datasets/HardNet++.pthz hardnet++znhttps://github.com/DagnyT/hardnet/raw/master/pretrained/train_liberty_with_aug/checkpoint_liberty_with_aug.pth liberty_augz/http://cmp.felk.cvut.cz/~mishkdmy/hardnet8v2.pt hardnet8v2cs^eZdZdZdZdeddfdd Zedej e ej d d d Z ej ej d d dZ Z S)HardNeta}Module, which computes HardNet descriptors of given grayscale patches of 32x32. This is based on the original code from paper "Working hard to know your neighbor's margins: Local descriptor learning loss". See :cite:`HardNet2017` for more details. Args: pretrained: Download and set pretrained weights to the model. Returns: torch.Tensor: HardNet descriptor of the patches. Shape: - Input: :math:`(B, 1, 32, 32)` - Output: :math:`(B, 128)` Examples: >>> input = torch.rand(16, 1, 32, 32) >>> hardnet = HardNet() >>> descs = hardnet(input) # 16x128 FN) pretrainedreturncsTtttjddddddtjdddttjddddddtjdddttjddddddd tjdddttjddddddtjdddttjdd ddddd tjd ddttjd d ddddtjd ddttd tjd d d dd tjd dd|_|rHdd}t j j t d|d}|j |ddd|dS)Nr F kernel_sizepaddingbiasZaffine@rZstriderr333333?rrcSs|SNZstoragelocrr_/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/kornia/feature/hardnet.pyDz"HardNet.__init__..rZ map_locationZ state_dictTstrict)super__init__nn SequentialConv2d BatchNorm2dReLUDropoutfeaturestorchhubload_state_dict_from_urlrload_state_dictevalselfr Z storage_fcnZpretrained_dict __class__rrr%(s:        zHardNet.__init__ư>xepsr cCs,tj|ddd\}}||||Sz4Utility function that normalizes the input by batch.)T)dimZkeepdimr-Zstd_meandetachr8r9spmprrr_normalize_inputIszHardNet._normalize_inputinputr cCsBt|gd||}||}||dd}tj|ddSN)B132rJrr=r )r>)rrDr,viewsizeF normalize)r3rFx_norm x_featuresZx_outrrrforwardRs   zHardNet.forward)F)r6)__name__ __module__ __qualname____doc__ patch_sizeboolr% staticmethodr-TensorfloatrDrQ __classcell__rrr4rrs !rcsheZdZdZdZdedfdd ZeddZede j e e j d d d Z e j e j d ddZ ZS)HardNet8aTModule, which computes HardNet8 descriptors of given grayscale patches of 32x32. This is based on the original code from paper "Improving the HardNet Descriptor". See :cite:`HardNet2020` for more details. Args: pretrained: Download and set pretrained weights to the model. Returns: torch.Tensor: HardNet8 descriptor of the patches. Shape: - Input: :math:`(B, 1, 32, 32)` - Output: :math:`(B, 128)` Examples: >>> input = torch.rand(16, 1, 32, 32) >>> hardnet = HardNet8() >>> descs = hardnet(input) # 16x128 r F)r cstttjddddddtjdddttjddddddtjdddttjddddddd tjdddttjddddddtjdddttjdd ddddd tjd ddttjd d ddddtjd ddttjd d ddddtjd ddttd tjd d dddtjd dd|_|j |j | dt j d d t jd| dt jd t jd|rdd}t jjtd|d}|j|dd|dS)Nr r r Frrrrrrrirr components)ZdtypemeancSs|Srrrrrrrr z#HardNet8.__init__..rr!Tr")r$r%r&r'r(r)r*r+r,apply weights_initZregister_bufferr-ZonesrZzerosr.r/rr0r1r2r4rrr%qsF         zHardNet8.__init__cCs@t|tjrr7cCs,tj|ddd\}}||||Sr:r?rArrrrDszHardNet8._normalize_inputrEcCs|t|gd||}||}tjtj|j}tjtj|j}t | | dd}t |||}t j |ddSrG)rrDr,r-ZjitZannotaterYr_r^rMrNrKrLmm)r3rFrOrPr_r^Zx_prePCAZpcarrrrQs  zHardNet8.forward)F)rg)rRrSrTrUrVrWr%rXrar-rYrZrDrQr[rrr4rr\Zs' r\)typingrrr-Ztorch.nnr&Ztorch.nn.functionalZ functionalrMZkornia.testingrrstr__annotations__Modulerr\rrrrs  I