o پi@sdZddlZddlmZmZddlmZddlmZm Z m Z m Z m Z m Z mZddlZddlmZddlmZmZddlmZmZmZmZmZmZddlmZmZmZmZd d l m!Z!d d l"m#Z#d d l$m%Z%m&Z&d d l'm(Z(m)Z)m*Z*ddgZ+eGdddZ,de-de.de.fddZ/ dde e.de e-de e.de-dee e.e e.ff ddZ0 dde-de.d e-d!e.d"e.d#e.dee e.e.e e.ffd$d%Z1  &dd'e.d(e.d)e.d*e.d+e.d,e e d-e2dej3fd.d/Z4  &dd'e.d(e.d)e.d*e.d+e.d,e e d-e2dej5fd0d1Z6 2  &dd3e e7d'e.d(e.d)e.d*e.d+ee.e.fd,e e d-e2de ej3fd4d5Z8Gd6d7d7ej3Z9Gd8d9d9ej3Z:Gd:d;d;ej3Z;Gdej3d?e7d@e2ddfdAdBZ=dCe e7efde e7effdDdEZ>e?didFe,dGdHdIddJdKdLe,dGdMdNdOdPdKdQe,dGdMdNdOdPdRdSdTe,dUdVdWdGdOdKdXe,dYdZd[dOdOdKd\e,d]d^d_dGd`dKdae,dbdcd_dUdddKdee,dfdgdhdidjdKdke,dldmdndYdodKdpe,d]dqdrdsdjdKdte,dudvdwdxdydKdze,d{d|d}d~dPdKde,ddddudjdKde,dGdHdIddJddde,dUddddOddde,dUdddOdddde,dYdddOdddde,dYdddOdddRdde,dUdddGdddde,d]dddGdddde,dfddWddPddde,dxdddddddde,ddddYdoddde,ddddYdoddRdde,dudvdwdxdyddde,ddddxd`ddde,dddddddde,dddddddde,dddddddde,dddddddde,dfddWddPddeedOddde,dPdfddWdddddčde,dddxddddddddǍ de,ddOdddddddddd΍ de,ddUdddddddddd΍ de,ddUdddddddddd΍ Z@de7de2deeDdd?dd4d5d@dAeDddBdd4d5d@dCeDddDdd4d5d@dEeCdddFeCdddGeCdddHeCdddIeCdddJeCdddKeCdddLeCddidMeCdddNeCdddOeCdddPeCdddQeCdddReCdddSeCdddTeCdddUeCdddVeCdddWeCdddXeCdddYeCdddZeCddd[eCddd\eCddZEe)dde2de RGB as most PyTorch models are * removed training specific dict entries from checkpoints and keep model state_dict only * remap names to match the ones here Supports weight loading from torchvision and classy-vision (incl VISSL SEER) A number of custom timm model definitions additions including: * stochastic depth, gradient checkpointing, layer-decay, configurable dilation * a pre-activation 'V' variant * only known RegNet-Z model definitions with pretrained weights Hacked together by / Copyright 2020 Ross Wightman N) dataclassreplace)partial)AnyCallableDictListOptionalUnionTupleIMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STD)ClassifierHead AvgPool2dSame ConvNormActSEModuleDropPath GroupNormAct) get_act_layerget_norm_act_layer create_conv2dmake_divisible)build_model_with_cfg)feature_take_indices)checkpoint_seq named_apply)generate_default_cfgsregister_modelregister_model_deprecationsRegNet RegNetCfgc@seZdZUdZdZeed<dZeed<dZe ed<dZ e ed <d Z eed <d Z e ed <dZ e ed<dZe ed<dZeed<dZeeed<dZeed<dZeed<dZeed<dZeeefed<dZeeefed<dS)r"z"RegNet architecture configuration.depthPw0q= ףPE@waHzG@wm group_size? bottle_ratiose_ratiogroup_min_ratio stem_widthconv1x1 downsampleF linear_outpreactr num_featuresrelu act_layer batchnorm norm_layerN)__name__ __module__ __qualname____doc__r$int__annotations__r&r(floatr*r,r.r0r1r3r5r strr6boolr7r8r:r rr<rFrFF/home/ubuntu/.local/lib/python3.10/site-packages/timm/models/regnet.pyr"-s"             fqreturncCstt|||S)zConverts a float to the closest non-zero int divisible by q. Args: f: Input float value. q: Quantization divisor. Returns: Quantized integer value. )rAround)rHrIrFrFrGquantize_floatAs rLr/widths bottle_ratiosgroups min_ratiocsvddt||D}ddt||D}r#fddt||D}n ddt||D}ddt||D}||fS)a,Adjusts the compatibility of widths and groups. Args: widths: List of channel widths. bottle_ratios: List of bottleneck ratios. groups: List of group sizes. min_ratio: Minimum ratio for divisibility. Returns: Tuple of adjusted widths and groups. cSsg|] \}}t||qSrFrA).0wbrFrFrG _z-adjust_widths_groups_comp..cSg|] \}}t||qSrF)min)rRgw_botrFrFrGrU`csg|] \}}t||qSrF)rrRrZrYrPrFrGrUcrVcSrWrF)rLr\rFrFrGrUer[cSsg|] \}}t||qSrFrQ)rRrZrTrFrFrGrUfrV)zip)rMrNrOrPbottleneck_widthsrFr]rGadjust_widths_groups_compNsr` width_slope width_initial width_multr$r,quantc s|dkr|dkr|dkr||dksJtj|tjd||}tt||t|}t|t||||}tt|t | d} } tj fddt | Dtjd} | | | fS)auGenerates per block widths from RegNet parameters. Args: width_slope: Slope parameter for width progression. width_initial: Initial width. width_mult: Width multiplier. depth: Network depth. group_size: Group convolution size. quant: Quantization factor. Returns: Tuple of (widths, num_stages, groups). rr)dtypecsg|]}qSrFrFrR_r,rFrGrUsz#generate_regnet..)torcharangefloat32rKlogmathpowlenuniquerAmaxitemtensorrangeint32tolist) rbrcrdr$r,re widths_cont width_expsrM num_stages max_stagerOrFrirGgenerate_regnetjs($"r|Fin_chsout_chs kernel_sizestridedilationr<r7c Cs^|ptj}|dkr|dkrdn|}|dkr|nd}|r$t|||||dSt||||||ddS)amCreate convolutional downsampling module. Args: in_chs: Input channels. out_chs: Output channels. kernel_size: Convolution kernel size. stride: Convolution stride. dilation: Convolution dilation. norm_layer: Normalization layer. preact: Use pre-activation. Returns: Downsampling module. r)rrF)rrr< apply_act)nn BatchNorm2drr)r}r~rrrr<r7rFrFrGdownsample_convs( rc Cs|ptj}|dkr |nd}t}|dks|dkr.|dkr#|dkr#tntj} | d|ddd}|r9t||ddd} n t||dd|dd} tj|| gS)aCreate average pool downsampling module. AvgPool Downsampling as in 'D' ResNet variants. This is not in RegNet space but I might experiment. Args: in_chs: Input channels. out_chs: Output channels. kernel_size: Convolution kernel size. stride: Convolution stride. dilation: Convolution dilation. norm_layer: Normalization layer. preact: Use pre-activation. Returns: Sequential downsampling module. rTF) ceil_modecount_include_padr)rr<r)rrIdentityr AvgPool2drr Sequential) r}r~rrrr<r7 avg_stridepool avg_pool_fnconvrFrFrGdownsample_avgs rrrdownsample_typec Cs|dvsJ||ks|dks|d|dkrr?r@rReLUrrAr rCrDrErrrrjTensorr __classcell__rFrFrrGrsP       Drcr) PreBottleneckznPre-activation RegNet Bottleneck block. Similar to Bottleneck but with pre-activation normalization. rrrr4FNr/r}r~rrr.r,r0r5r6r:r<rc stt|t| | }tt||}||}|||_t||dd|_|||_ t||d||d|d|_ |rJtt||}t ||| d|_ nt |_ |||_t||dd|_t|||d||dd|_| dkrst| |_d St |_d S) aInitialize pre-activation RegNet Bottleneck block. Args: in_chs: Input channels. out_chs: Output channels. stride: Convolution stride. dilation: Dilation rates for conv2 and shortcut. bottle_ratio: Bottleneck ratio (reduction factor). group_size: Group convolution size. se_ratio: Squeeze-and-excitation ratio. downsample: Shortcut downsampling type. linear_out: Use linear activation for output. act_layer: Activation layer. norm_layer: Normalization layer. drop_block: Drop block layer. drop_path_rate: Stochastic depth drop rate. r)rrr)rrrrOrT)rrrr7N)rrrrrArKnorm1rrnorm2rrrrrnorm3rrr5rr)rr}r~rrr.r,r0r5r6r:r<rrnorm_act_layerrrOrrrFrGrps>!     $ zPreBottleneck.__init__rJcCsdS)z?Zero-initialize the last batch norm (no-op for pre-activation).NrFrrFrFrGrszPreBottleneck.zero_init_lastrcCsl||}|}||}||}||}||}||}||}|jdur4||||}|Sr) rrrrrrrr5rrrFrFrGrs        zPreBottleneck.forwardrrrFrFrrGrjsP       CrcsbeZdZdZdefdedededededeeed e ffd d Z d e j d e j fddZ ZS)RegStagezRegNet stage (sequence of blocks with the same output shape). A stage consists of multiple bottleneck blocks with the same output dimensions. Nr$r}r~rrdrop_path_ratesblock_fnc stt|d|_|dvrdnd} t|D]:} | dkr|nd} | dkr&|n|} | |f} |dur4|| nd}d| d}|||| |f| | |d ||} qdS) aInitialize RegNet stage. Args: depth: Number of blocks in stage. in_chs: Input channels. out_chs: Output channels. stride: Stride for first block. dilation: Dilation rate. drop_path_rates: Drop path rates for each block. block_fn: Block class to use. **block_kwargs: Additional block arguments. F)rrrrrNr/zb{})rrr)rrrgrad_checkpointingruformat add_module)rr$r}r~rrrr block_kwargsfirst_dilationi block_stride block_in_chsblock_dilationdprnamerrFrGrs0  zRegStage.__init__rrJcCs<|jrtjst||}|S|D]}||}q|S)zForward pass through all blocks in the stage. Args: x: Input tensor. Returns: Output tensor. )rrjjit is_scriptingrchildren)rrblockrFrFrGrs   zRegStage.forward)r=r>r?r@rrAr rrCrrrjrrrrFrFrrGrs(  .rcseZdZdZ       d8ded ed ed ed ed ededeffdd Z   d9deded edede e e ee fe ee fff ddZejjd:dede ee ffddZejjd;deddfddZejjdejfdd Zdd%eee efd-ed.ede efd/d0Zd$ejdejfd1d2Zd:d$ejd3edejfd4d5Zd$ejdejfd6d7ZZ S)?r!zRegNet-X, Y, and Z Models. Paper: https://arxiv.org/abs/2003.13678 Original Impl: https://github.com/facebookresearch/pycls/blob/master/pycls/models/regnet.py rr2rr/Tcfgin_chans num_classes output_stride global_pool drop_raterrc  st||_||_|dvsJt|fi| }|j} t|j|jd} |j r1t || ddd|_ n t || dfddi| |_ t| dddg|_ | } d} |j|||d \}}t|d ks]J|j rbtnt}t|D]2\}}d |d }||td| |d |||d} | |d9} |j t| | |dg7_ qh|jrt | |jfdd i| |_|j|_n|jp|j }|rt|jnt|_| |_|j|_t|j|||d|_tt t!|d|dS)aInitialize RegNet model. Args: cfg: Model architecture configuration. in_chans: Number of input channels. num_classes: Number of classifier classes. output_stride: Output stride of network, one of (8, 16, 32). global_pool: Global pooling type. drop_rate: Dropout rate. drop_path_rate: Stochastic depth drop-path rate. zero_init_last: Zero-init last weight of residual path. kwargs: Extra kwargs overlayed onto cfg. )rar2rrrrrstem)num_chs reductionmodule)rrzs{}r)r}rr~r) in_featuresr pool_typer)rNrF)"rrrrrr3rr:r<r7rrr feature_info_get_stage_argsrprr enumeraterrrr8 final_convr6rrrhead_hidden_sizerheadrr _init_weights)rrrrrrrrrkwargsr3na_args prev_width curr_strideper_stage_args common_argsrr stage_args stage_name final_actrrFrGrsd      zRegNet.__init__rdefault_striderJc sdtjjjjj\}}}tjt|dd\}} | | }} fddt |D} g} g} d} d}t |D]}| |krI||9}d}n|}| |9} | || |q.rrrN)dimcSsg|]}|qSrF)rw)rRrrFrFrGrUsr])r~rrr$r.r,rcsg|] }tt|qSrF)rr^)rRparams) arg_namesrFrGrUs)r5r0r6r:r<)r|r(r&r*r$r,rjrqrtrwruappendlinspacesumcumsum tensor_splitr`r1r^rr5r0r6r:r<)rrrrrrMrzstage_gs stage_widths stage_depthsstage_br stage_stridesstage_dilations net_striderrhr dpr_tensor split_indices stage_dprrrrF)rrrGrnsF       zRegNet._get_stage_argsFcoarsecCstd|rddSddS)z"Group parameters for optimization.z^stemz^s(\d+)z^s(\d+)\.b(\d+))rblocks)r)rr rFrFrG group_matchers zRegNet.group_matcherenableNcCs$t|ddD]}||_q dS)z)Enable or disable gradient checkpointing.rrN)listrr)rrsrFrFrGset_grad_checkpointingszRegNet.set_grad_checkpointingcCs|jjS)zGet the classifier head.)rfcrrFrFrGget_classifierszRegNet.get_classifiercCs||_|jj||ddS)zReset the classifier head. Args: num_classes: Number of classes for new classifier. global_pool: Global pooling type. )rN)rrreset)rrrrFrFrGreset_classifierszRegNet.reset_classifierNCHWrindicesnorm stop_early output_fmtintermediates_onlyc Cs|dvsJdg}td|\}} d} ||}| |vr!||d} |r+| d| } | D]} | d7} t|| |}| |vrC||q-|rH|S| dkrQ||}||fS) a Forward features that returns intermediates. Args: x: Input image tensor indices: Take last n blocks if int, all if None, select matching indices if sequence norm: Apply norm layer to compatible intermediates stop_early: Stop iterating over blocks when last desired intermediate hit output_fmt: Shape of intermediate feature outputs intermediates_only: Only return intermediate features Returns: )rzOutput shape must be NCHW.rs1s2s3s4Nrr)rrrgetattrr) rrrrrrr intermediates take_indices max_indexfeat_idx layer_namesnrFrFrGforward_intermediatess*     zRegNet.forward_intermediatesr prune_norm prune_headcCs^td|\}}d}||d}|D] }t||tq|dkr%t|_|r-|dd|S)aEPrune layers not required for specified intermediates. Args: indices: Indices of intermediate layers to keep. prune_norm: Whether to prune normalization layer. prune_head: Whether to prune the classifier head. Returns: List of indices that were kept. rrNrrr)rsetattrrrrr)rrr+r,r%r&r(r)rFrFrGprune_intermediate_layerss   z RegNet.prune_intermediate_layerscCs@||}||}||}||}||}||}|S)zForward pass through feature extraction layers. Args: x: Input tensor. Returns: Feature tensor. )rrr r!r"rrrrFrFrGforward_featuress     zRegNet.forward_features pre_logitscCs|r |j||dS||S)zForward pass through classifier head. Args: x: Input features. pre_logits: Return features before final linear layer. Returns: Classification logits or features. )r1)r)rrr1rFrFrG forward_head#s zRegNet.forward_headcCs||}||}|S)zoForward pass. Args: x: Input tensor. Returns: Output logits. )r0r2r/rFrFrGr/s zRegNet.forward)rrr2rr/r/T)rr2r/F)T)N)NFFrF)rFT)!r=r>r?r@r"rArDrCrErr rrrrrjrignorerrrModulerr rrr r*r.r0r2rrrFrFrrGr!s  V <   2  rrrrcCst|tjr4|jd|jd|j}||j}|jjdt d||j dur2|j j dSdSt|tj rTtjj|jddd|j durRtj|j dSdS|rat|drc|dSdSdS) zInitialize module weights. Args: module: PyTorch module to initialize. name: Module name. zero_init_last: Zero-initialize last layer weights. rr@Nr/g{Gz?)meanstdr) isinstancerConv2dr out_channelsrOrdatanormal_rnsqrtbiaszero_Linearrrhasattrr)rrrfan_outrFrFrGr=s      r state_dictcCsn|d|}gd}d|vrtddl}|ddd}i}|dD]1\}}|dd }|d d }|d d d|}|dd|}|D] \}}|||}qD|||<q"|dD]\}}d|vsfd|vrgqZ|dd}|||<qZ|Sd|vrddl}i}|D]0\}}|dd }|dd }|ddd|}|D] \}}|||}q|dd}|||<q|S|S)zFilter and remap state dict keys for compatibility. Args: state_dict: Raw state dictionary. Returns: Filtered state dictionary. model))zf.a.0z conv1.conv)zf.a.1zconv1.bn)zf.b.0z conv2.conv)zf.b.1zconv2.bn)z f.final_bnconv3.bn)zf.se.excitation.0zse.fc1)zf.se.excitation.2zse.fc2)zf.ser)zf.c.0 conv3.conv)zf.c.1rF)zf.crG)zproj.0downsample.conv)zproj.1z downsample.bn)projrHclassy_state_dictrN base_modeltrunkz_feature_blocks.conv1.stem.0 stem.convz_feature_blocks.conv1.stem.1zstem.bnz&^_feature_blocks.res\d.block(\d)-(\d+)cS(dt|ddt|ddS)Nrr.brrAgrouprrFrFrGw(z_filter_fn..zs(\d)\.b(\d+)\.bnzs\1.b\2.downsample.bnheadsprojection_head prototypesz0.clf.0head.fcz stem.0.weightzstem.0zstem.1z)trunk_output.block(\d)\.block(\d+)\-(\d+)cSrN)NrrrOrrPrRrFrFrGrSrTzfc.zhead.fc.)getreitemsrsub)rDreplacesrZoutkvrrrFrFrG _filter_fnSsN            rb regnetx_002r+gQ8B@gQ@ )r&r(r*r,r$ regnetx_004g{Gz8@gRQ@rregnetx_004_tvg?)r&r(r*r,r$r1 regnetx_0060g\(|B@gQ@ regnetx_0088g= ףpA@g= ףp=@ regnetx_016r%gzGA@g@ regnetx_032Xg(\O:@ regnetx_040`g33333SC@gq= ףp@( regnetx_064g ףp=jN@g(\@ regnetx_080gHzH@g ףp= @x regnetx_120gףp= WR@g(\@p regnetx_160gQK@g@ regnetx_320@gףp= wQ@r6 regnety_002r)r&r(r*r,r$r0 regnety_004gp= ;@gQ@ regnety_006gQE@@g(\@ regnety_008gQkC@g333333@regnety_008_tv)r&r(r*r,r$r0r1 regnety_016g(\µ4@g333333@ regnety_032r'r)r# regnety_040g)\h?@@ regnety_064g\(@@g)\(@H regnety_080gGz4S@gQ@regnety_080_tv regnety_120 regnety_160gQZ@gףp= @ regnety_320g)\\@g= ףp=@ regnety_640i`g(\ob@iH regnety_1280ig(\d@g)\(@i regnety_2560ig(\l@iuregnety_040_sgnsiluri)r&r(r*r,r$r0r:r< regnetv_040T)r$r&r(r*r,r0r7r: regnetv_064r) r$r&r(r*r,r0r7r:r5 regnetz_005gffffff%@gGz@rg@i) r$r&r(r*r,r.r0r5r6r8r: regnetz_040g-@g+@ regnetz_040_hivariant pretrainedcKstt||ft|td|S)zCreate a RegNet model. Args: variant: Model variant name. pretrained: Load pretrained weights. **kwargs: Additional model arguments. Returns: RegNet model instance. ) model_cfgpretrained_filter_fn)rr! model_cfgsrb)rrrrFrFrG_create_regnets rurlc Ks"|dddddddttdd d |S) zCreate default configuration dictionary. Args: url: Model weight URL. **kwargs: Additional configuration options. Returns: Configuration dictionary. rrrr)r rgffffff?r-bicubicrMrX) rr input_size pool_sizetest_input_sizecrop_pct test_crop_pct interpolationr7r8 first_conv classifierr rrrFrFrG_cfgs rc K"|dddddttdddd d |S) zCreate pycls configuration dictionary. Args: url: Model weight URL. **kwargs: Additional configuration options. Returns: Configuration dictionary. rrrg?rrMrXmitz)https://github.com/facebookresearch/pycls rrrrrrr7r8rrlicense origin_urlr rrFrFrG_cfgpyc rc Kr) zCreate torchvision v2 configuration dictionary. Args: url: Model weight URL. **kwargs: Additional configuration options. Returns: Configuration dictionary. rrrgzG?rrMrXz bsd-3-clausez!https://github.com/pytorch/visionrr rrFrFrG_cfgtv2 rrzregnety_032.ra_in1kztimm/znhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/regnety_032_ra-7f2439f9.pth) hf_hub_idrzregnety_040.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnety_040_ra3-670e1166.pthzregnety_064.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnety_064_ra3-aa26dc7d.pthzregnety_080.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnety_080_ra3-1fdc4344.pthzregnety_120.sw_in12k_ft_in1k)rzregnety_160.sw_in12k_ft_in1kzregnety_160.lion_in12k_ft_in1kzregnety_120.sw_in12ki-.)rrzregnety_160.sw_in12kzregnety_040_sgn.untrained)rzregnetv_040.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetv_040_ra3-c248f51f.pthr)rrrzregnetv_064.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetv_064_ra3-530616c2.pthzregnetz_005.untrainedzregnetz_040.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetz_040_ra3-9007edf5.pth)rr)rarar-)rrr)rrrrrrzregnetz_040_h.ra3_in1kzthttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetz_040h_ra3-f594343b.pthzregnety_160.deit_in1kzhttps://download.pytorch.org/models/regnet_x_16gf-ba3796d7.pthzregnetx_320.tv2_in1kz>https://download.pytorch.org/models/regnet_x_32gf-6eb8fdc6.pthzregnety_004.tv2_in1kz?https://download.pytorch.org/models/regnet_y_400mf-e6988f5f.pthzregnety_008_tv.tv2_in1kz?https://download.pytorch.org/models/regnet_y_800mf-58fc7688.pthzregnety_016.tv2_in1kz?https://download.pytorch.org/models/regnet_y_1_6gf-0d7bc02a.pthzregnety_032.tv2_in1kz?https://download.pytorch.org/models/regnet_y_3_2gf-9180c971.pthzregnety_080_tv.tv2_in1kz=https://download.pytorch.org/models/regnet_y_8gf-dc2b1b54.pthzregnety_160.tv2_in1kz>https://download.pytorch.org/models/regnet_y_16gf-3e4a00f9.pthzregnety_320.tv2_in1kz>https://download.pytorch.org/models/regnet_y_32gf-8db6d4b5.pthzregnety_160.swag_ft_in1kzChttps://download.pytorch.org/models/regnet_y_16gf_swag-43afe44d.pthz cc-by-nc-4.0)rr) r)rrrrrrzregnety_320.swag_ft_in1kzChttps://download.pytorch.org/models/regnet_y_32gf_swag-04fdfa75.pthzregnety_1280.swag_ft_in1kzDhttps://download.pytorch.org/models/regnet_y_128gf_swag-c8ce3e52.pthzregnety_160.swag_lc_in1kzFhttps://download.pytorch.org/models/regnet_y_16gf_lc_swag-f3ec0043.pth)rrrzregnety_320.swag_lc_in1kzFhttps://download.pytorch.org/models/regnet_y_32gf_lc_swag-e1583746.pthzregnety_1280.swag_lc_in1kzGhttps://download.pytorch.org/models/regnet_y_128gf_lc_swag-cbe8ce12.pthzregnety_320.seer_ft_in1kotherz)https://github.com/facebookresearch/visslzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet32_finetuned_in1k_model_final_checkpoint_phase78.torch)rrrrrrrzregnety_640.seer_ft_in1kzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet64_finetuned_in1k_model_final_checkpoint_phase78.torchzregnety_1280.seer_ft_in1kzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet128_finetuned_in1k_model_final_checkpoint_phase78.torchzregnety_2560.seer_ft_in1kzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet256_finetuned_in1k_model_final_checkpoint_phase38.torchzregnety_320.seerzihttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_regnet32d/seer_regnet32gf_model_iteration244000.torch)rrrrrzregnety_640.seerzphttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_regnet64/seer_regnet64gf_model_final_checkpoint_phase0.torchzregnety_1280.seerzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/swav_ig1b_regnet128Gf_cnstant_bs32_node16_sinkhorn10_proto16k_syncBN64_warmup8k/model_final_checkpoint_phase0.torchzregnetx_002.pycls_in1kzregnetx_004.pycls_in1kzregnetx_006.pycls_in1kzregnetx_008.pycls_in1kzregnetx_016.pycls_in1kzregnetx_032.pycls_in1kzregnetx_040.pycls_in1kzregnetx_064.pycls_in1kzregnetx_080.pycls_in1kzregnetx_120.pycls_in1kzregnetx_160.pycls_in1kzregnetx_320.pycls_in1kzregnety_002.pycls_in1kzregnety_004.pycls_in1kzregnety_006.pycls_in1kzregnety_008.pycls_in1kzregnety_016.pycls_in1kzregnety_032.pycls_in1kzregnety_040.pycls_in1kzregnety_064.pycls_in1kzregnety_080.pycls_in1kzregnety_120.pycls_in1kzregnety_160.pycls_in1kzregnety_320.pycls_in1kcKtd|fi|S)z RegNetX-200MFrcrrrrFrFrGrccKr)z RegNetX-400MFrerrrFrFrGrercKr)z+RegNetX-400MF w/ torchvision group roundingrgrrrFrFrGrgrcKr)z RegNetX-600MFrhrrrFrFrGrhrcKr)z RegNetX-800MFrjrrrFrFrGrjrcKr)z RegNetX-1.6GFrlrrrFrFrGrlrcKr)z RegNetX-3.2GFrnrrrFrFrGrnrcKr)z RegNetX-4.0GFrqrrrFrFrGrqrcKr)z RegNetX-6.4GFrurrrFrFrGrurcKr)z RegNetX-8.0GFrxrrrFrFrGrxrcKr)z RegNetX-12GFrzrrrFrFrGrzrcKr)z RegNetX-16GFr~rrrFrFrGr~rcKr)z RegNetX-32GFrrrrFrFrGr rcKr)z RegNetY-200MFrrrrFrFrGr&rcKr)z RegNetY-400MFrrrrFrFrGr,rcKr)z RegNetY-600MFrrrrFrFrGr2rcKr)z RegNetY-800MFrrrrFrFrGr8rcKr)z+RegNetY-800MF w/ torchvision group roundingrrrrFrFrGr>rcKr)z RegNetY-1.6GFrrrrFrFrGrDrcKr)z RegNetY-3.2GFrrrrFrFrGrJrcKr)z RegNetY-4.0GFrrrrFrFrGrPrcKr)z RegNetY-6.4GFrrrrFrFrGrVrcKr)z RegNetY-8.0GFrrrrFrFrGr\rcKr)z+RegNetY-8.0GF w/ torchvision group roundingrrrrFrFrGrbrcKr)z RegNetY-12GFrrrrFrFrGrhrcKr)z RegNetY-16GFrrrrFrFrGrnrcKr)z RegNetY-32GFrrrrFrFrGrtrcKr)z RegNetY-64GFrrrrFrFrGrzrcKr)z RegNetY-128GFrrrrFrFrGrrcKr)z RegNetY-256GFrrrrFrFrGrrcKr)zRegNetY-4.0GF w/ GroupNorm rrrrFrFrGrrcKr)zRegNetV-4.0GF (pre-activation)rrrrFrFrGrrcKr)zRegNetV-6.4GF (pre-activation)rrrrFrFrGrrcKtd|fddi|S)zRegNetZ-500MF NOTE: config found in https://github.com/facebookresearch/ClassyVision/blob/main/classy_vision/models/regnet.py but it's not clear it is equivalent to paper model as not detailed in the paper. rrFrrrFrFrGrcKr)RegNetZ-4.0GF NOTE: config found in https://github.com/facebookresearch/ClassyVision/blob/main/classy_vision/models/regnet.py but it's not clear it is equivalent to paper model as not detailed in the paper. rrFrrrFrFrGrrcKr)rrrFrrrFrFrGrr regnetz_040h)r/)ra)rrrNF)rNF)rFrF)rr3)kr@rn dataclassesrr functoolsrtypingrrrrr r r rjtorch.nnr timm.datar r timm.layersrrrrrrrrrr_builderr _featuresr _manipulaterr _registryrrr __all__r"rCrArLr`r|rEr5rrrrDrrrrr!rrbrrrrrr default_cfgsrcrergrhrjrlrnrqrurxrzr~rrrrrrrrrrrrrrrrrrrrrrrrr=rFrFrFrGs $     " $ 1 ,  &eeE +" B      !%  *,159@       %& * 0369<?BEILORUX[_ c g lorv {            ! " # $ % & ' ( * + , - . / 0 1 2 3 4 59