hh , dZddlmcmZddlmZddlmZddlm Z ddl m Z ddl m Z ddl mZdd lmZd Zid d d ddddddddddddddddddd d!d"d#d$d%d&d'd(d)d*d+d,d-d.d/d0d1d2d3d4Zid gd5gd6d7d8d9d:d gd;gd<d=d8d9d:dgd>gd?d@d8d9d:dgdAgdBd=d8d9d:dgdCgdDd@d8d9d:dgdEgdFdGd8d9d:dgdHgdIdJd8d9d:dgdKgdLdMd8d9d:dgdNgdOdPd8d9d:dgdQgdRdSd8d9d:dgdTgdUdVd8d9d:d!gdWgdXdYd8d9d:d#gd5gd6d7d8dZd:d%gd[gd\d7d8dZd:d'gd]gd^d=d8dZd:d)gd_gd`d=d8dZd:d+gdagdbd@d8dZd:gdcgddd@d8dZd:gdegdfdgd8dZd:gdhgdidjd8dZd:gdKgdkdMd8dZd:gdQgdRdSd8dZd:gdlgdmdSd8dZd:gdngdodpd8dZd:d4ZdqZddrZddsZddtZdduZ ddvZ ddwZddxZddyZ ddzZed{d| dd}Zed~d ddZedd ddZ edd ddZ!edd ddZ"edd ddZ#edd ddZ$edd ddZ%edd ddZ&edd ddZ'edd ddZ(edd ddZ)edd ddZ*edd ddZ+edd ddZ,edd ddZ-edd ddZ.edd ddZ/edd ddZ0edd ddZ1edd ddZ2edd ddZ3edd ddZ4edd dd„Z5ejmdìīe_ejmdŬīe_ejmdƬīe _ejmdǬīe!_ejmdȬīe"_ejmdɬīe#_ejmdʬīe$_ejmdˬīe%_ejmd̬īe&_ejmdͬīe'_ejmdάīe(_ejmdϬīe)_ejmdЬīe*_ejmdѬīe+_ejmdҬīe,_ejmdӬīe-_ejmdԬīe._ejmdլīe/_ejmd֬īe0_ejmd׬īe1_ejmdجīe2_ejmd٬īe3_ejmdڬīe4_ejmd۬īe5_edܫdd݄Z7edޫdd߄Z8e jpje8_y)zRegNet models for TF-Keras. References: - [Designing Network Design Spaces](https://arxiv.org/abs/2003.13678) (CVPR 2020) - [Fast and Accurate Model Scaling](https://arxiv.org/abs/2103.06877) (CVPR 2021) N)backend)layers)imagenet_utils)training) data_utils) layer_utils) keras_exportzDhttps://storage.googleapis.com/tensorflow/keras-applications/regnet/x002)@49fb46e56cde07fdaf57bffd851461a86548f6a3a4baef234dd37290b826c0b8@5445b66cd50445eb7ecab094c1e78d4d3d29375439d1a7798861c4af15ffff21x004)@3523c7f5ac0dbbcc2fd6d83b3570e7540f7449d3301cc22c29547302114e4088@de139bf07a66c9256f2277bf5c1b6dd2d5a3a891a5f8a925a10c8a0a113fd6f3x006)@340216ef334a7bae30daac9f414e693c136fac9ab868704bbfcc9ce6a5ec74bb@a43ec97ad62f86b2a96a783bfdc63a5a54de02eef54f26379ea05e1bf90a9505x008)@8f145d6a5fae6da62677bb8d26eb92d0b9dfe143ec1ebf68b24a57ae50a2763d@3c7e4b0917359304dc18e644475c5c1f5e88d795542b676439c4a3acd63b7207x016)@31c386f4c7bfef4c021a583099aa79c1b3928057ba1b7d182f174674c5ef3510@1b8e3d545d190271204a7b2165936a227d26b79bb7922bac5ee4d303091bf17ax032)@6c025df1409e5ea846375bc9dfa240956cca87ef57384d93fef7d6fa90ca8c7f@9cd4522806c0fcca01b37874188b2bd394d7c419956d77472a4e072b01d99041x040)@ba128046c588a26dbd3b3a011b26cb7fa3cf8f269c184c132372cb20b6eb54c1@b4ed0ca0b9a98e789e05000e830403a7ade4d8afa01c73491c44610195198afex064)@0f4489c3cd3ad979bd6b0324213998bcb36dc861d178f977997ebfe53c3ba564@3e706fa416a18dfda14c713423eba8041ae2509db3e0a611d5f599b5268a46c4x080)@76320e43272719df648db37271a247c22eb6e810fe469c37a5db7e2cb696d162@7b1ce8e29ceefec10a6569640ee329dba7fbc98b5d0f6346aabade058b66cf29x120)@5cafc461b78897d5e4f24e68cb406d18e75f31105ef620e7682b611bb355eb3a@36174ddd0299db04a42631d028abcb1cc7afec2b705e42bd28fcd325e5d596bfx160)@8093f57a5824b181fb734ea21ae34b1f7ee42c5298e63cf6d587c290973195d2@9d1485050bdf19531ffa1ed7827c75850e0f2972118a996b91aa9264b088fd43x320)@91fb3e6f4e9e44b3687e80977f7f4412ee9937c0c704232664fc83e4322ea01e@9db7eacc37b85c98184070e1a172e6104c00846f44bcd4e727da9e50d9692398y002)@1e8091c674532b1a61c04f6393a9c570113e0197f22bd1b98cc4c4fe800c6465@f63221f63d625b8e201221499682587bfe29d33f50a4c4f4d53be00f66c0f12cy004)@752fdbad21c78911bf1dcb8c513e5a0e14697b068e5d9e73525dbaa416d18d8e@45e6ba8309a17a77e67afc05228454b2e0ee6be0dae65edc0f31f1da10cc066by006)@98942e07b273da500ff9699a1f88aca78dfad4375faabb0bab784bb0dace80a9@b70261cba4e60013c99d130cc098d2fce629ff978a445663b6fa4f8fc099a2bey008)@1b099377cc9a4fb183159a6f9b24bc998e5659d25a449f40c90cbffcbcfdcae4@b11f5432a216ee640fe9be6e32939defa8d08b8d136349bf3690715a98752ca1y016)@b7ce1f5e223f0941c960602de922bcf846288ce7a4c33b2a4f2e4ac4b480045b@d7404f50205e82d793e219afb9eb2bfeb781b6b2d316a6128c6d7d7dacab7f57)@6a6a545cf3549973554c9b94f0cd40e25f229fffb1e7f7ac779a59dcbee612bd@eb3ac1c45ec60f4f031c3f5180573422b1cf7bebc26c004637517372f68f8937)@98d00118b335162bbffe8f1329e54e5c8e75ee09b2a5414f97b0ddfc56e796f6@b5be2a5e5f072ecdd9c0b8a437cd896df0efa1f6a1f77e41caa8719b7dfcb05d)@65c948c7a18aaecaad2d1bd4fd978987425604ba6669ef55a1faa0069a2804b7@885c4b7ed7ea339daca7dafa1a62cb7d41b1068897ef90a5a3d71b4a2e2db31a)@7a2c62da2982e369a4984d3c7c3b32d6f8d3748a71cb37a31156c436c37f3e95@3d119577e1e3bf8d153b895e8ea9e4ec150ff2d92abdca711b6e949c3fd7115d)@a96ab0d27d3ae35a422ee7df0d789069b3e3217a99334e0ce861a96595bc5986@4a6fa387108380b730b71feea2ad80b5224b5ea9dc21dc156c93fe3c6186485c)@45067240ffbc7ca2591313fee2f80dbdda6d66ec1a7451446f9a6d00d8f7ac6e@ead1e6b568be8f34447ec8941299a9df4368736ba9a8205de5427fa20a1fb316)@b05e173e4ae635cfa22d06392ee3741284d17dadfee68f2aa6fd8cb2b7561112@cad78f74a586e24c61d38be17f3ae53bb9674380174d2585da1a526b8c20e1fd)y032y040y064y080y120y160y320)rR)8ipX)depthswidths group_width default_size block_type)rRrT ) @i)rRrT)0`irU)rRrfrTrg)rc )r`rS r`)Hii)r`r`)riiirh)r`rgr`)Prji0iP()r`rSrnrR)rpiiiXrV)r`rgrrrR)rurjiix)r`rg rR)rYip)r`rq rR)rzirk)r`rTr|rR)Prmi@i rpY)rRrfrqrq)rhhi)rRrfrTrS)rhr{r}i`)rRrfrXr`)rcrki@i)r`rqr`)rhrwrix)r`rgr|rR)ro@i)r`rqrar`)rkrsr~i@rc)r`rTrtr`)rlriro)rpryrzi)r`rSrxrR)rYryii )r`rgrarR)iipira Instantiates the {name} architecture. Reference: - [Designing Network Design Spaces](https://arxiv.org/abs/2003.13678) (CVPR 2020) For image classification use cases, see [this page for detailed examples]( https://keras.io/api/applications/#usage-examples-for-image-classification-models). For transfer learning use cases, make sure to read the [guide to transfer learning & fine-tuning]( https://keras.io/guides/transfer_learning/). Note: Each TF-Keras Application expects a specific kind of input preprocessing. For Regnets, preprocessing is included in the model using a `Rescaling` layer. RegNet models expect their inputs to be float or uint8 tensors of pixels with values in the [0-255] range. The naming of models is as follows: `RegNet` where `block_type` is one of `(X, Y)` and `flops` signifies hundred million floating point operations. For example RegNetY064 corresponds to RegNet with Y block and 6.4 giga flops (64 hundred million flops). Args: include_top: Whether to include the fully-connected layer at the top of the network. Defaults to `True`. weights: One of `None` (random initialization), `"imagenet"` (pre-training on ImageNet), or the path to the weights file to be loaded. Defaults to `"imagenet"`. input_tensor: Optional TF-Keras tensor (i.e. output of `layers.Input()`) to use as image input for the model. input_shape: Optional shape tuple, only to be specified if `include_top` is False. It should have exactly 3 inputs channels. pooling: Optional pooling mode for feature extraction when `include_top` is `False`. - `None` means that the output of the model will be the 4D tensor output of the last convolutional layer. - `avg` means that global average pooling will be applied to the output of the last convolutional layer, and thus the output of the model will be a 2D tensor. - `max` means that global max pooling will be applied. Defaults to `None`. classes: Optional number of classes to classify images into, only to be specified if `include_top` is True, and if no `weights` argument is specified. 1000 is how many ImageNet classes there are. Defaults to `1000`. classifier_activation: A `str` or callable. The activation function to use on the "top" layer. Ignored unless `include_top=True`. Set `classifier_activation=None` to return the logits of the "top" layer. When loading pretrained weights, `classifier_activation` can only be `None` or `"softmax"`. Defaults to `"softmax"`. Returns: A `keras.Model` instance. cX!dttjdzfd}|S)zRescales and normalizes inputs to [0,1] and ImageNet mean and std. Args: name: name prefix Returns: Rescaled and normalized tensor prestemcHtjddz|}|S)Ngp?_prestem_rescaling)scalename)r Rescalingxrs ^/var/www/html/dev/engine/venv/lib/python3.12/site-packages/tf_keras/src/applications/regnet.pyapplyzPreStem..applys1 F  D+?$?  strrget_uidrrs` rPreStemr~s. |3wy9:: LrcX!dttjdzfd}|S)zImplementation of RegNet stem. (Common to all model variants) Args: name: name prefix Returns: Output tensor of the Stem stemc tjdddddddz|}tjd d d z |}tjd z|}|S)Nrbrfrfr`Fsame he_normal _stem_conv)stridesuse_biaspaddingkernel_initializerr?h㈵>_stem_bnmomentumepsilonr _stem_relur)rConv2DBatchNormalizationReLUrs rrzStem..applys FMM  * $    F % %$TJ->   2FKKTL0 1! 4rrrs` rStemrs. |GOOF344 LrcZttjdfd}|S)zImplements the Squeeze & Excite block (https://arxiv.org/abs/1709.01507). Args: filters_in: input filters to the block se_filters: filters to squeeze to name: name prefix Returns: A function object squeeze_and_excitec tjdzd|}tjddddz|}tjdd dd z|}tjj ||}|S) N_squeeze_and_excite_gapT)rkeepdimsrRrRrelur_squeeze_and_excite_squeeze) activationrrsigmoid_squeeze_and_excite_excite)rGlobalAveragePooling2Drtfmathmultiply)inputsr filters_inr se_filterss rrz$SqueezeAndExciteBlock..applys F ) )11D   FMM  *55    FMM   *44    GG  Q 'rr)rrrrs``` rSqueezeAndExciteBlockrs* |7??#789* Lrcbttjdfd}|S)aYImplementation of X Block. Reference: [Designing Network Design Spaces](https://arxiv.org/abs/2003.13678) Args: filters_in: filters in the input tensor filters_out: filters in the output tensor group_width: group width stride: stride name: name prefix Returns: Output tensor of the block xblockc Nk7rdk(rtddddd z}dk7rFtjddd d z |}tjd d dz|}n|}tjddd dz|}tjd d dz|}tjdz|}tjdd|dd dz|}tjd d dz|}tjdz|}tjddd dz|}tjd d dz|}tjdz||z}|SNrRInput filters() and output filters(z) are not equal for stride 4. Input and output filters must be equal for stride=.rFr _skip_1x1)rrrrrr_skip_bnr _conv_1x1_1rrr_conv_1x1_1_bn_conv_1x1_1_relurrr _conv_3x3rrgroupsrrr _conv_3x3_bn_conv_3x3_relu _conv_1x1_2_conv_1x1_2_bn _exit_relu) ValueErrorrrrr) rrskiprr filters_outr]rstrides rrzXBlock..applys1  $1  -&-(,,284,,2816  + Q;6==#.K' D6,,d 1BDD FMM  * %     F % %$T4D-D   8FKKT$66 7 : FMM  * #    F % %$TN-B   6FKKT$44 5a 8 FMM  * %    F % %$T4D-D   2FKKTL0 1!d( ;rr)rrr]rrrs````` rXBlockrs1 |7??8,-GGR Lrcfttjdfd}|S)aImplementation of Y Block. Reference: [Designing Network Design Spaces](https://arxiv.org/abs/2003.13678) Args: filters_in: filters in the input tensor filters_out: filters in the output tensor group_width: group width stride: stride squeeze_excite_ratio: expansion ration for Squeeze and Excite block name: name prefix Returns: Output tensor of the block yblockc k7r dk(rtddd d d z}t z} dk7rFtjd dd d z |}tjd d dz|}n|}tjddd dz|}tjd d dz|}tj dz|}tjdd |dd dz|}tjd d dz|}tj dz|}t ||}tjddd dz|}tjd d dz|}tj dz||z}|Sr)rintrrrrr) rrrrrrrr]rsqueeze_excite_ratiors rrzYBlock..applyOsZ  $1  -&-(,,284,,2816  +&::; Q;6==#.K' D6,,d 1BDD FMM  * %     F % %$T4D-D   8FKKT$66 7 : FMM  * #    F % %$TN-B   6FKKT$44 5a 8 F !+z Ea H FMM  * %    F % %$T4D-D   2FKKTL0 1!d( ;rr)rrr]rrrrs`````` rYBlockr6s1, |7??8,-JJX Lrcjttjdfd}|S)aImplementation of Z block Reference: [Fast and Accurate Model Scaling](https://arxiv.org/abs/2103.06877). Args: filters_in: filters in the input tensor filters_out: filters in the output tensor group_width: group width stride: stride squeeze_excite_ratio: expansion ration for Squeeze and Excite block bottleneck_ratio: inverted bottleneck ratio name: name prefix Returns: Output tensor of the block zblockc k7r dk(rtddd d d z}t z}tz }tj|ddd d z |}tjd d dz|}t j j|}tj|dd |dd dz|}tjd d dz|}t j j|}t|| }tjddd dz |}tjd d dz|} dk7r|S||zS)NrRrrz)are not equal for stride rrrFrrrrrrrrrrrrrrr) rrrrrrnnsilur) rrrinv_btlneck_filtersrbottleneck_ratiorrr]rrrs rrzZBlock..applys  $1  ,A+O,,284,,2816   +&::; !+0@"@A FMM  * %     F % %$T4D-D   EEJJqM FMM  * #    F % %$TN-B   EEJJqM ""5z M FMM  * %    F % %$T4D-D   Q;Hv: rr)rrr]rrrrrs``````` rZBlockrs/. |7??8,-==~ Lrcfttjdfd}|S)auImplementation of Stage in RegNet. Args: block_type: must be one of "X", "Y", "Z" depth: depth of stage, number of blocks to use group_width: group width of all blocks in this stage filters_in: input filters to this stage filters_out: output filters from this stage name: name prefix Returns: Output tensor of Stage stagec |}dk(rFtdd|}tdD]}td||}|Sdk(rFtdd z|}tdD]}td ||}|Sd k(rFtdd |}tdD]}td ||}|St dd)NrZr` _XBlock_0)rrrR_XBlock_rr _YBlock_0_YBlock_Z _ZBlock_0_ZBlock_z Block type `z<` not recognized.block_type must be one of (`X`, `Y`, `Z`). )rrangerrNotImplementedError) rrir_depthrrr]rs rrzStage..apply s   vY'   A1e_ F 6!-    TG3 K'   A1e_ F 6!-    6)3 vY'   A1e_ F 6!-     &zl+>> rr)r_rr]rrrrs`````` rStagers/ |7??7+,44l LrcVttjdfd}|S)zImplementation of classification head of RegNet. Args: num_classes: number of classes for Dense layer name: name prefix Returns: Classification head function. headctjdz|}tjdz|}|S)N _head_gapr head_dense)rrDense)rr num_classess rrzHead..applyQs@ BF ) )tk/A B1 E ?FLL4,+> ? Brr)rrrs`` rHeadrDs) |7??6*+ Lrc |dvs4tjjj|s t d|dk(r|r| dk7r t dt j | |dtj||} | tj| }n/tj| stj| | }n| }| tj| d }n|}|}|rt| |}t| |}d}t!d D]4}||}||}t#||||||dzt%|z |}|}6|r)t'| |}t j(| |n?| dk(rtj*|}n| dk(rtj,|}t/j0|||}|dk(r]|rd}t2|ddd }nd}t2|ddd}||z}t5j6|t8|zd|}|j;||S||j;||S)a Instantiates RegNet architecture given specific configuration. Args: depths: An iterable containing depths for each individual stages. widths: An iterable containing output channel width of each individual stages group_width: Number of channels to be used in each group. See grouped convolutions for more information. block_type: Must be one of `{"X", "Y", "Z"}`. For more details see the papers "Designing network design spaces" and "Fast and Accurate Model Scaling" default_size: Default input image size. model_name: An optional name for the model. include_preprocessing: boolean denoting whther to include preprocessing in the model include_top: Boolean denoting whether to include classification head to the model. weights: one of `None` (random initialization), "imagenet" (pre-training on ImageNet), or the path to the weights file to be loaded. input_tensor: optional TF-Keras tensor (i.e. output of `layers.Input()`) to use as image input for the model. input_shape: optional shape tuple, only to be specified if `include_top` is False. It should have exactly 3 inputs channels. pooling: optional pooling mode for feature extraction when `include_top` is `False`. - `None` means that the output of the model will be the 4D tensor output of the last convolutional layer. - `avg` means that global average pooling will be applied to the output of the last convolutional layer, and thus the output of the model will be a 2D tensor. - `max` means that global max pooling will be applied. classes: optional number of classes to classify images into, only to be specified if `include_top` is True, and if no `weights` argument is specified. classifier_activation: A `str` or callable. The activation function to use on the "top" layer. Ignored unless `include_top=True`. Set `classifier_activation=None` to return the logits of the "top" layer. Returns: A `keras.Model` instance. Raises: ValueError: in case of invalid argument for `weights`, or invalid input shape. ValueError: if `classifier_activation` is not `softmax` or `None` when using a pretrained top layer. ValueError: if `include_top` is True but `num_classes` is not 1000. ValueError: if `block_type` is not one of `{"X", "Y", "Z"}` >NimagenetzThe `weights` argument should be either `None` (random initialization), `imagenet` (pre-training on ImageNet), or the path to the weights file to be loaded.rzWIf using `weights` as `'imagenet'` with `include_top` as true, `classes` should be 1000rb)r^min_size data_formatrequire_flattenweightsN)shape)tensorr rrrS_Stage_)ravgmax)routputsrz.h5z _notop.h5rRmodels) cache_subdir file_hash)riogfileexistsrrobtain_input_shaperimage_data_formatrInputis_keras_tensorrget_source_inputsrrrrrrvalidate_activationrGlobalMaxPooling2DrModelWEIGHTS_HASHESrget_fileBASE_WEIGHTS_PATH load_weights)r[r\r]r_r^ model_nameinclude_preprocessing include_topr  input_tensor input_shapepoolingclassesclassifier_activation img_inputrr in_channels num_stager out_channelsmodel file_suffixr file_name weights_paths rRegNetr3Ys@ ) )RUU[[-?-?-H <  *D 1  !33!--/# KLL{3 &&|4 L LI$I..|rs ""! 4()*:Ka  a    a  a  a"  #a*  +a2  3a:  ;aB  CaJ  KaR  SaZ  [ab  caj  kar  saz  {aB  CaJ       {aRi $  i $  i $  i, %  -i: %  ;iH &  IiV &  Wid '  eir &  si@ '  AiN '  Oi\ (  ]ij $  kix %  yiF %  GiT %  Uib %  cir &   '   '   '   '   (   (  Ei V<~*@#LZB  eX  YxGT6   #Y~*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:*,K   #:$** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= #** *= :; < *<==>=,>>FFr