:j. F UddlZddlmZddlmZddlmZmZmZddl Z ddl m Z m Z ddl m Z mZddlmZmZdd lmZd d lmZmZmZd d lmZd d lmZmZmZgdZGdde ZGdde j@Z!Gdde jDZ#Gdde j@Z$GddZ%Gdde jDZ&de%deede'dede&f d Z(d!ed"Z)e*e+efe,d#<ie)d$d%d&Z-Gd'd(eZ.Gd)d*eZ/Gd+d,eZ0Gd-d.eZ1Gd/d0eZ2Gd1d2eZ3Gd3d4eZ4Gd5d6eZ5Gd7d8eZ6Gd9d:eZ7Gd;deZ9Gd?d@eZ:GdAdBeZ;GdCdDeZ<eedEe.jzfFddGdHdee.de'dede&fdIZ>eedEe/jzfFddGdHdee/de'dede&fdJZ?eedEe0jzfFddGdHdee0de'dede&fdKZ@eedEe1jzfFddGdHdee1de'dede&fdLZAeedEe2jzfFddGdHdee2de'dede&fdMZBeedEe3jzfFddGdHdee3de'dede&fdNZCeedEe4jzfFddGdHdee4de'dede&fdOZDeedPFddGdHdee5de'dede&fdQZEeedEe6jzfFddGdHdee6de'dede&fdRZFeedEe7jzfFddGdHdee7de'dede&fdSZGeedEe8jzfFddGdHdee8de'dede&fdTZHeedEe9jzfFddGdHdee9de'dede&fdUZIeedEe:jzfFddGdHdee:de'dede&fdVZJeedEe;jzfFddGdHdee;de'dede&fdWZKeedEe2     rHr8ceZdZdZdededededejfdedejfded ed e ed d ffd Z xZ S)BottleneckTransformz/Bottleneck transformation: 1x1, 3x3 [+SE], 1x1.r9r:rBr;.r< group_widthbottleneck_multiplierse_ratior=Nc >t} tt||z} | |z} t|| dd||| d<t| | d|| ||| d<|r(tt||z} t | | || d<t| |dd|d| d <t || y) Nrr@ar?)rArBgroupsr;r<b)input_channelssqueeze_channels activationsec)rrMroundr r rCrD)rEr9r:rBr;r<rSrTrUlayersw_bg width_se_outrFs rGrDzBottleneckTransform.__init__Fs/:m% $99:; ; * cqz\l s + !F1fv s  uX%89:L,"!-+F4L + ! ]a s   rH rIrJrKrLrMrrrNfloatrrDrOrPs@rGrRrRCs9#!#!#! #! S"))^, #! #3 >2 #!#! %#!5/#! #!#!rHrRceZdZdZ ddededededejfdedejfd ed ed e ed dffd Z de d e fdZ xZ S)ResBottleneckBlockz>Residual bottleneck block: x + F(x), F = bottleneck transform.Nr9r:rBr;.r<rSrTrUr=c t |d|_||k7xs|dk7} | rt||d||d|_t |||||||||_|d|_y)Nrr@T)inplace)rCrDprojr rRfr\) rEr9r:rBr;r<rSrTrU should_projrFs rGrDzResBottleneckBlock.__init__os}  9,>&A+ ,)6jkoDI%       !   +48rHxc|j$|j||j|z}n||j|z}|j|SN)rjrkr\rErms rGforwardzResBottleneckBlock.forwardsF 99  ! tvvay(ADFF1I Aq!!rH)r?N)rIrJrKrLrMrrrNrerrDr rqrOrPs@rGrgrglsH'*$(999 9 S"))^, 9 #3 >2 99 %95/9 9@""F"rHrgceZdZdZ ddedededededejfd edejfd edejfd ed ed e ededdffd Z xZ S)AnyStagez;AnyNet stage (sequence of blocks w/ the same output shape).Nr9r:rBdepthblock_constructor.r;r<rSrTrU stage_indexr=c t|t|D]7} || dk(r|n||| dk(r|nd|||| | } |jd| d| | 9y)Nrrblock-)rCrDrange add_module)rEr9r:rBrurvr;r<rSrTrUrwiryrFs rGrDzAnyStage.__init__sr u >A%F q&a % E OOeK=!5u = >rH)NrrdrPs@rGrtrtsE%)>>> >  > $CN3 >S"))^,>#3 >2>> %>5/>> >>rHrtceZdZ ddeedeedeedeedeedeeddfd Ze dd ed ed ed edededeede ddfdZ dZ e deedeedeede eeeeffdZy) BlockParamsNdepthswidths group_widthsbottleneck_multipliersstridesrUr=cX||_||_||_||_||_||_yrorrrrrrU)rErrrrrrUs rGrDzBlockParams.__init__s0  (&<#   rHruw_0w_aw_mrSrTkwargsc d} d} |dks|dks |dks|dzdk7r tdtj||z|z} tjtj| |z t j|z } tjtj |tj|| z| | zjj} tt| }t| dgzdg| z| dgzdg| z}|Dcgc]\}}}}||k7xs||k7}}}}}t| |ddDcgc] \}}|s | }}}tjtjt|Dcgc] \}}|s | c}}jj}| g|z}|g|z}|g|z}|j!|||\}}|||||||Scc}}}}wcc}}wcc}}w) a) Programmatically compute all the per-block settings, given the RegNet parameters. The first step is to compute the quantized linear block parameters, in log space. Key parameters are: - `w_a` is the width progression slope - `w_0` is the initial width - `w_m` is the width stepping in the log space In other terms `log(block_width) = log(w_0) + w_m * block_capacity`, with `bock_capacity` ramping up following the w_0 and w_a params. This block width is finally quantized to multiples of 8. The second step is to compute the parameters per stage, taking into account the skip connection and the final 1x1 convolutions. We use the fact that the output width is constant within a stage. r rrzInvalid RegNet settingsNr) ValueErrortorcharanger_logmathdividepowrMtolistlensetzipdifftensor enumerate"_adjust_widths_groups_compatibilty)clsrurrrrSrTrUrQUANTSTRIDE widths_contblock_capacity block_widths num_stages split_helperwwprrpsplitst stage_widthsd stage_depthsrrrs rGfrom_init_paramszBlockParams.from_init_paramss@ 7cQh#(cAgl67 7ll5)C/#5 UYY{S/@%ADHHSM%QR ELLuyyn7U1UW\$]^affkkmttv \*+  A3  C,  A3  C,   :FFFB2!r'$Q"W$FF&),s &DJdaJ Jzz%,,i>O/UdaST/U"VW[[]ddf (Z'"7!8:!E#}z1 &)%K%K 0,& " l%#9   GJ/UsG5 G=G= HHct|j|j|j|j|j Sro)rrrrrr)rEs rG_get_expanded_paramsz BlockParams._get_expanded_paramss-4;; dkk4;L;LdNiNijjrHrbottleneck_ratiosct||Dcgc]\}}t||z}}}t||Dcgc]\}}t||}}}t||Dcgc]\}}t||} }}t| |Dcgc]\}}t||z }}}||fScc}}wcc}}wcc}}wcc}}w)zl Adjusts the compatibility of widths and groups, depending on the bottleneck ratio. )rrMminr) rrrrrYrrbw_botgroup_widths_minws_bots rGrz.BlockParams._adjust_widths_groups_compatibiltys*-\;L)MNA#a!e*NN:=lF:SThaC5MTT=@HX++F3 )) KP 'A!RYY'--*Q]]1-==Ns #-8PQAr~~. ahh'qvv&Aryy)s=qvv& 'rHrmc|j|}|j|}|j|}|jd}|j |}|S)Nr) start_dim)rrrflattenrrps rGrqzRegNet.forwardxsN IIaL   a  LLO III " GGAJrH)i NNNN) rIrJrKrrMrrrrNrDr rqrOrPs@rGrr(s 8<9=9=9=M'!M'M' M' HS"))^45 M' Xc299n56 M'Xc299n56M'Xc299n56M' M'^FrHrrweightsprogressrr=c |#t|dt|jd|jdt t j dd}t|fd|i|}|"|j|j|d|S) Nr categoriesr;gh㈵>g?)epsmomentumT)r check_hash) rrmetapoprrrrload_state_dictget_state_dict)rrrrr;models rG_regnetrs fmSl9S5TUL'"..eVY*Z[J < AJ A& AE g44hSW4XY LrHr)min_sizer _COMMON_METAz(https://github.com/facebookresearch/SWAGz:https://github.com/facebookresearch/SWAG/blob/main/LICENSE)recipelicensec eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)rz?https://download.pytorch.org/models/regnet_y_400mf-c65dace8.pth crop_sizeiPIBRhttps://github.com/pytorch/vision/tree/main/references/classification#small-models ImageNet-1KgmR@gV@zacc@1zacc@5g|?5^?guV0@XThese weights reproduce closely the results of the paper using a simple training recipe. num_paramsr_metrics_ops _file_size_docsurl transformsrz?https://download.pytorch.org/models/regnet_y_400mf-e6988f5f.pthr resize_size8https://github.com/pytorch/vision/issues/3995#new-recipeg~jtR@gsh|/W@$ These weights improve upon the results of the original paper by using a modified version of TorchVision's `new training recipe `_. N rIrJrKrrr r IMAGENET1K_V1 IMAGENET1K_V2DEFAULTrrHrGrr M.#>    !j##   s  M$ M.#3O  !P##    M,GrHrc eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)rz?https://download.pytorch.org/models/regnet_y_800mf-1b27b58c.pthrri'brrg{GS@gbX9HW@rg㥛 ?g/$8@rrrz?https://download.pytorch.org/models/regnet_y_800mf-58fc7688.pthrr r gES@g㥛 W@r Nr rrHrGrrrrHrc eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)rz?https://download.pytorch.org/models/regnet_y_1_6gf-b11a554e.pthrri~rrg|S@g}W@rgn?g~jtE@rrrz?https://download.pytorch.org/models/regnet_y_1_6gf-0d7bc02a.pthrr r gMb8T@g#~jW@r Nr rrHrGrrs M.#>    "j##   s  M$ M.#3O  "P##    M,GrHrc eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)rz?https://download.pytorch.org/models/regnet_y_3_2gf-b5a9779c.pthrri2(Shttps://github.com/pytorch/vision/tree/main/references/classification#medium-modelsrg&1S@g/ݤW@rg rh @g?5^IR@rrrz?https://download.pytorch.org/models/regnet_y_3_2gf-9180c971.pthrr r g+~T@g|?5W@r Nr rrHrGrr% M.#>    "k##   s  M$ M.#3O  "P##    M,GrHrc eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)rz=https://download.pytorch.org/models/regnet_y_8gf-d0d0e4a8.pthrriXrrg5^I T@gPnW@rgV- @gOnb@rrrz=https://download.pytorch.org/models/regnet_y_8gf-dc2b1b54.pthrr r gET@gQX@r Nr rrHrGrrQ K.#>    "k##  !s  M$ K.#3O  "P##  ! M,GrHrc XeZdZedeediedddddd id d d d Zedeeddiedddddd id d dd Zedeedde jie ddddd idd ddZ edeedde jie dddd d!d id d d"d#Z eZy$)%rz>https://download.pytorch.org/models/regnet_y_16gf-9e6ed7dd.pthrri{Rhttps://github.com/pytorch/vision/tree/main/references/classification#large-modelsrgB`"T@g(\W@rgm/@gp= s@rrrz>https://download.pytorch.org/models/regnet_y_16gf-3e4a00f9.pthrr r gbX9T@gEX@r zChttps://download.pytorch.org/models/regnet_y_16gf_swag-43afe44d.pthrr  interpolationgT㥛ĀU@g~jtX@gGz^G@ These weights are learnt via transfer learning by end-to-end fine-tuning the original `SWAG `_ weights on ImageNet-1K data. rrrrrzFhttps://download.pytorch.org/models/regnet_y_16gf_lc_swag-f3ec0043.pth+https://github.com/pytorch/vision/pull/5793gX9vT@gV-OX@ These weights are composed of the original frozen `SWAG `_ trunk weights and a linear classifier learnt on top of them trained on ImageNet-1K data. rrrrrrNrIrJrKrrr rrrrBICUBIC_COMMON_SWAG_METAIMAGENET1K_SWAG_E2E_V1IMAGENET1K_SWAG_LINEAR_V1rrrHrGrr}sh L.#>    "j##   s  M$ L.#3O  "P##    M,% Q 3CO`OhOh   "##    ,!( T 3CO`OhOh   C"##    !.GrHrc XeZdZedeediedddddd id d d d Zedeeddiedddddd id d dd Zedeedde jie ddddd idd ddZ edeedde jie dddd d!d id d d"d#Z eZy$)%rz>https://download.pytorch.org/models/regnet_y_32gf-4dee3f7a.pthrri<rrgx&18T@g(\W@rgp= #@@gS㥛P@rrrz>https://download.pytorch.org/models/regnet_y_32gf-8db6d4b5.pthrr r gd;OT@gZd;X@r zChttps://download.pytorch.org/models/regnet_y_32gf_swag-04fdfa75.pthrrgʡU@gI +X@g/ݴW@rrzFhttps://download.pytorch.org/models/regnet_y_32gf_lc_swag-e1583746.pthr g+'U@gQ^X@r!r"Nr#rrHrGrrsh L.#>    #j##  !s  M$ L.#3O  #P##  ! M,% Q 3CO`OhOh   ###  ! ,!( T 3CO`OhOh   C###  ! !.GrHrc eZdZedeeddejiedddddid d d d  Z edeeddejieddddddidd dd Z e Z y)r zDhttps://download.pytorch.org/models/regnet_y_128gf_swag-c8ce3e52.pthrri^o&rg;OV@gS㥫X@rgQiw@g㥛 ;@rrrzGhttps://download.pytorch.org/models/regnet_y_128gf_lc_swag-cbe8ce12.pthrr g1ZU@gtvX@gx&_@r!r"N) rIrJrKrrr rr$r%r&r'rrrHrGr r /s$ R 3CO`OhOh   ###  " ,!( U 3CO`OhOh   C###  " !.%GrHr c eZdZedeediedddddd id d d d Zedeeddiedddddd id ddd ZeZ y)r!z?https://download.pytorch.org/models/regnet_x_400mf-adf1edd5.pthrriSrrgL7A`5R@g̼V@rg"~?g5^I B5@rrrz?https://download.pytorch.org/models/regnet_x_400mf-62229a5f.pthrr Dhttps://github.com/pytorch/vision/issues/3995#new-recipe-with-fixresgKR@gS㥛W@goA5@r Nr rrHrGr!r!` M.#>    !j##   s  M$ M.#3O  !\##    M,GrHr!c eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)r"z?https://download.pytorch.org/models/regnet_x_800mf-ad17e45c.pthrrinrrg!rhR@gʡEW@rg?gR;@rrrz?https://download.pytorch.org/models/regnet_x_800mf-94a99ebd.pthrr r+g rhaS@g/tW@r Nr rrHrGr"r"s M.#>    !j##   s  M$ M.#3O  !\##    M,GrHr"c eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)r#z?https://download.pytorch.org/models/regnet_x_1_6gf-e3633e7f.pthrri:rrg(\BS@g\(\W@rgS?gZdA@rrrz?https://download.pytorch.org/models/regnet_x_1_6gf-a12f2b72.pthrr r+gnS@g^I W@r Nr rrHrGr#r#r,rHr#c eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)r$z?https://download.pytorch.org/models/regnet_x_3_2gf-f342aeae.pthrri(hrrgKS@gsh|W@rg"~j @gT㥛`M@rrrz?https://download.pytorch.org/models/regnet_x_3_2gf-7071aa85.pthrr r gCLT@gQW@r Nr rrHrGr$r$rrHr$c eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)r%z=https://download.pytorch.org/models/regnet_x_8gf-03ceed89.pthrri[rrgtS@gClW@rg{G@g;Ob@rrrz=https://download.pytorch.org/models/regnet_x_8gf-2b70d774.pthrr r gSkT@gZdW@r Nr rrHrGr%r%rrHr%c eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)r&z>https://download.pytorch.org/models/regnet_x_16gf-2007eb11.pthrri9<rrgʡET@g#~jW@rgo/@gMbi@rrrz>https://download.pytorch.org/models/regnet_x_16gf-ba3796d7.pthrr r gҭT@gC X@r Nr rrHrGr&r&<s L.#>    "k##  !s  M$ L.#3O  "P##  ! M,GrHr&c eZdZedeediedddddd id d d d Zedeeddiedddddd id d dd ZeZ y)r'z>https://download.pytorch.org/models/regnet_x_32gf-9d47f8d0.pthrrimrrg+'T@gZd;W@rg#~j?@gvy@rrrz>https://download.pytorch.org/models/regnet_x_32gf-6eb8fdc6.pthrr r g7A`T@gOnX@r Nr rrHrGr'r'hs L.#>    #j##  !s  M$ L.#3O  #P##  ! M,GrHr' pretrained)rT)rrc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_400MF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_400MF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_400MF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_400MF_Weights :members: 0gp= ;@gQ@r?rurrrrSrUr)rverifyrrrrrrparamss rGr(r(sM(%++G4G  ) ) x4]^im xqw xF 67H 7 77rHc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_800MF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_800MF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_800MF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_800MF_Weights :members: 8gQkC@g333333@r5r7r8r)rr9rrrr:s rGr)r)sM(%++G4G  ) ) x3\^im xqw xF 67H 7 77rHc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_1.6GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_1_6GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_1_6GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_1_6GF_Weights :members: r6g(\µ4@g333333@r7r8r)rr9rrrr:s rGr*r*T(%++G4G  ) )be2PVF 67H 7 77rHc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_3.2GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_3_2GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_3_2GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_3_2GF_Weights :members: Pgq= ףPE@gHzG@rAr7r8r)rr9rrrr:s rGr+r+rBrHc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_8GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_8GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_8GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_8GF_Weights :members: gGz4S@gQ@r>r7r8r)rr9rrrr:s rGr,r,sT(#))'2G  ) )cu$BQWF 67H 7 77rHc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_16GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_16GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_16GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_16GF_Weights :members: gQZ@gףp= @pr7r8r)rr9rrrr:s rGr-r-T($**73G  ) )cv4S4SYF 67H 7 77rHc tj|}tjdddddddd|}t |||fi|S)a Constructs a RegNetY_32GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_32GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_32GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_32GF_Weights :members: rg)\\@g= ףp=@r7r8r)rr9rrrr:s rGr.r.8rMrH)r3Nc tj|}tjdddddddd|}t |||fi|S) a Constructs a RegNetY_128GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_Y_128GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_Y_128GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_Y_128GF_Weights :members: r@ig(\d@g)\(@ir7r8r)r r9rrrr:s rGr/r/TsT(%++G4G  ) )cv4S4SYF 67H 7 77rHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_400MF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_400MF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_400MF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_400MF_Weights :members: rAg{Gz8@gRQ@r5rurrrrSr)r!r9rrrr:s rGr0r0pJ(%++G4G  ) ) j4]_ jci jF 67H 7 77rHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_800MF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_800MF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_800MF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_800MF_Weights :members: r5r>g= ףpA@g= ףp=@rSr)r"r9rrrr:s rGr1r1rTrHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_1.6GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_1_6GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_1_6GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_1_6GF_Weights :members: rJrEgzGA@@rArSr)r#r9rrrr:s rGr2r2rTrHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_3.2GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_3_2GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_3_2GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_3_2GF_Weights :members: Xg(\O:@rWr6rSr)r$r9rrrr:s rGr3r3rTrHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_8GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_8GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_8GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_8GF_Weights :members: rEgHzH@g ףp= @xrSr)r%r9rrrr:s rGr4r4sJ(#))'2G  ) ) k4]` kdj kF 67H 7 77rHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_16GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_16GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_16GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_16GF_Weights :members: rRgQK@g@rSr)r&r9rrrr:s rGr5r5J($**73G  ) ) kC]` kdj kF 67H 7 77rHc ~tj|}tjddddddd|}t |||fi|S)a Constructs a RegNetX_32GF architecture from `Designing Network Design Spaces `_. Args: weights (:class:`~torchvision.models.RegNet_X_32GF_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.RegNet_X_32GF_Weights` below for more details and possible values. By default, no pretrained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to either ``torchvision.models.regnet.RegNet`` or ``torchvision.models.regnet.BlockParams`` class. Please refer to the `source code `_ for more detail about the classes. .. autoclass:: torchvision.models.RegNet_X_32GF_Weights :members: r\i@gףp= wQ@rrSr)r'r9rrrr:s rGr6r6 rarH)Mr collectionsr functoolsrtypingrrrrrr ops.miscr r transforms._presetsr rutilsr_apirrr_metar_utilsrrr__all__r8rrRrNrgrtrrboolrrdictstr__annotations__r%rrrrrrrr r!r"r#r$r%r&r'rr(r)r*r+r,r-r.r/r0r1r2r3r4r5r6rrHrGrrs( #** >H'66'SS F  '  &!"--&!R("("V>r}}>Dl.l.^XRYYXv k "   &&  d38n 8K)[)X)[)X)[)X)[)X);)XVKVrVKVr.%[.%b)[)X)[)X)[)X)[)X);)X)K)X)K)X,0F0T0T!UVBFY]8x(>?8RV8il8qw8W80,0F0T0T!UVBFY]8x(>?8RV8il8qw8W80,0F0T0T!UVBFY]8x(>?8RV8il8qw8W84,0F0T0T!UVBFY]8x(>?8RV8il8qw8W84,0D0R0R!ST>BUY8X&:;8d8eh8ms8U84,0E0S0S!TU@DW[8h'<=8PT8gj8ou8V84,0E0S0S!TU@DW[8h'<=8PT8gj8ou8V84!56BFY]8x(>?8RV8il8qw8784,0F0T0T!UVBFY]8x(>?8RV8il8qw8W80,0F0T0T!UVBFY]8x(>?8RV8il8qw8W80,0F0T0T!UVBFY]8x(>?8RV8il8qw8W80,0F0T0T!UVBFY]8x(>?8RV8il8qw8W80,0D0R0R!ST>BUY8X&:;8d8eh8ms8U80,0E0S0S!TU@DW[8h'<=8PT8gj8ou8V80,0E0S0S!TU@DW[8h'<=8PT8gj8ou8V8rH