h SSKJr SSKJrJrJr SSKrSSKJr SSKJ r SSK J r SSK J r S S KJrJrJr S S KJr S S KJrJr /S Qr"SS\R05r"SS\R45r"SS\R45r"SS\5r\"5\"S\R<4S9SSS.S\\S\S\S\4Sjj55r g))partial)AnyCallableOptionalN)nn)Conv3dNormActivation)VideoClassification)_log_api_usage_once)register_modelWeights WeightsEnum)_KINETICS400_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface)S3D S3D_Weightss3dcb^\rSrSrS\S\S\S\S\S\S\R44 U4S jjrS r U=r $) TemporalSeparableConv in_planes out_planes kernel_sizestridepadding norm_layer.c x>[TU][UUSX34SXD4SXU4SUS9[UUUSS4USS4USS4SUS95 g)NrF)rrrbiasr)super__init__r)selfrrrrrr __class__s T/var/www/auris/envauris/lib/python3.13/site-packages/torchvision/models/video/s3d.pyr#TemporalSeparableConv.__init__sp   96*G-%  !(!Q/1~ !Q%  ) __name__ __module__ __qualname____firstlineno__intrrModuler#__static_attributes__ __classcell__r%s@r&rrsT         S"))^,  r(rcp^\rSrSrS\S\S\S\S\S\S\S \S \R44U4S jjrS r S r U=r $)SepInceptionBlock3D6rb0_outb1_midb1_outb2_midb2_outb3_outr.c |>[T U]5 [XSSUS9Ul[R "[XSSUS9[ X4SSSUS95Ul[R "[XSSUS9[ XVSSSUS95Ul[R "[R"SSSS9[XSSUS95Ul g)Nr rrrr )rrrrr r r rrr) r"r#rbranch0r Sequentialrbranch1branch2 MaxPool3dbranch3) r$rr6r7r8r9r:r;rr%s r&r#SepInceptionBlock3D.__init__7s +I1UVcmn }} !Xb c !&aSTak l  }} !Xb c !&aSTak l  }} LLYq! D !Xb c  r(cURU5nURU5nURU5nURU5n[R "X#XE4S5nU$)Nr )r@rBrCrEtorchcat)r$xx0x1x2x3outs r&forwardSepInceptionBlock3D.forwardRsP \\!_ \\!_ \\!_ \\!_ii(!, r()r@rBrCrE) r*r+r,r-r.rrr/r#rPr0r1r2s@r&r4r46sm           S"))^, 6r(r4c ^\rSrSrSrS S\S\S\\S\ RR4SS4U4S jjjr S r S rU=r$) r\a;S3D main class. Args: num_class (int): number of classes for the classification task. dropout (float): dropout probability. norm_layer (Optional[Callable]): Module specifying the normalization layer to use. Inputs: x (Tensor): batch of videos with dimensions (batch, channel, time, height, width) N num_classesdropoutr.returnc>[TU]5 [U5 Uc[[R SSS9n[R "[SSSSSU5[R"SSS S 9[SSS S US 9[SS SS S U5[R"SSS S 9[S SSSSSSU5[SSSS SSSU5[R"SSSS 9[SS SSSSSU5[SSSSSSSU5[SSSSSSSU5[SSSSSSSU5[S SSS!SSSU5[R"SSS"S 9[S#SSS!SSSU5[S#S$S S$SSSU55Ul [R"S%S S&9Ul [R "[R"US'9[R"S(US S S)S*95Ulg)+NgMbP?)epsmomentumr @r )r r r )r r r )rr r r?r r=` r>)r r r )r r r i0ipi ii@)rrri@i)r r[r[)rr)piT)rrr!)r"r#r rr BatchNorm3drArrDrr4features AvgPool3davgpoolDropoutConv3d classifier)r$rTrUrr%s r&r# S3D.__init__hs D!   UUKJ !!RAq* = LLYy) T %   ""c1aJ ? LLYy) T RS"b"j I S#sBB K LLYy) T S"c2r2z J S#sBB K S#sBB K S#sBB K S#sBS* M LLYy) T S#sBS* M S#sBS* M-  0|| !D -- JJ ! IIdKQqt L r(cURU5nURU5nURU5n[R"USS9nU$)N)r r )dim)rkrmrprHmean)r$rJs r&rP S3D.forwards@ MM!  LLO OOA  JJqi (r()rmrprk)ig?N)r*r+r,r-__doc__r.floatrrrHrr/r#rPr0r1r2s@r&rr\se ?C ( ( ( Xc588??&:;< (  ( ( Tr(rcR\rSrSr\"S\"\SSS9SS\SSS S S S S .0SSS. S9r\r Sr g)rz4https://download.pytorch.org/models/s3d-d76dad2f.pth)rfrf)rara) crop_size resize_sizezOhttps://github.com/pytorch/vision/tree/main/references/video_classification#s3dzThe weights aim to approximate the accuracy of the paper. The accuracies are estimated on clip-level with parameters `frame_rate=15`, `clips_per_video=1`, and `clip_len=128`.i0~z Kinetics-400gd;OQ@g33333V@)zacc@1zacc@5gv1@gF?@) min_sizemin_temporal_size categoriesrecipe_docs num_params_metrics_ops _file_size)url transformsmetar)N) r*r+r,r-rrr rKINETICS400_V1DEFAULTr0r)r(r&rrs` B  " #!#1g\"##!  # N6Gr(r pretrained)weightsT)rprogressrrkwargsrVc [RU5nUb#[US[URS55 [ S0UD6nUbUR URUSS95 U$)a_Construct Separable 3D CNN model. Reference: `Rethinking Spatiotemporal Feature Learning `__. .. betastatus:: video module Args: weights (:class:`~torchvision.models.video.S3D_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.video.S3D_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.video.S3D`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.video.S3D_Weights :members: rTrT)r check_hashr))rverifyrlenrrload_state_dictget_state_dict)rrrmodels r&rrsi0  )GfmSl9S5TU M&ME g44hSW4XY Lr()! functoolsrtypingrrrrHrtorchvision.ops.miscrtransforms._presetsr utilsr _apir rr_metar_utilsrr__all__rArr/r4rrrboolrr)r(r&rs** 56(77+C  BMM @#"))#L;"));|+>, 0J0J!KL,04 H[) D SV [^ M r(