6bi dSrSSKrSSKJs Jr SSKJr SSKJ r SSKJ r SSKJ r SSK J r SSKJr SS KJr SS KJr S rS S SSSS.r/SQ/SQSS./SQ/SQSS./SQ/SQSS./SQ/SQSS./SQ/SQSS.S.rSr"SS\ R05r"SS\ R05rS=S jrS>S!jrS?S"jrS@S#jr\"S$S%5SAS&j5r\"S'S(5SBS)j5r \"S*S+5SCS,j5r!\"S-S.5SDS/j5r"\"S0S15SES2j5r#\RIS3S49\l\RIS5S49\ l\RIS6S49\!l\RIS7S49\"l\RIS8S49\#l\"S95S>S:j5r%\"S;5SFS<j5r&\ RLR\&lg)GzxConvNeXt models for TF-Keras. References: - [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) (CVPR 2022) N)backend) initializers)layers)utils)imagenet_utils) sequential)training) keras_exportzFhttps://storage.googleapis.com/tensorflow/keras-applications/convnext/)@8ae6e78ce2933352b1ef4008e6dd2f17bc40771563877d156bc6426c7cf503ff@d547c096cabd03329d7be5562c5e14798aa39ed24b474157cef5e85ab9e49ef1)@ce1277d8f1ee5a0ef0e171469089c18f5233860ceaf9b168049cb9263fd7483c@6fc8009faa2f00c1c1dfce59feea9b0745eb260a7dd11bee65c8e20843da6eab)@52cbb006d3dadd03f6e095a8ca1aca47aecdd75acb4bc74bce1f5c695d0086e6@40a20c5548a5e9202f69735ecc06c990e6b7c9d2de39f0361e27baeb24cb7c45)@070c5ed9ed289581e477741d3b34beffa920db8cf590899d6d2c67fba2a198a6@96f02b6f0753d4f543261bc9d09bed650f24dd6bc02ddde3066135b63d23a1cd)@c1f5ccab661354fc3a79a10fa99af82f0fbf10ec65cb894a3ae0815f17a889ee@de3f8a54174130e0cecdc71583354753d557fcf1f4487331558e2a16ba0cfe05) convnext_tinyconvnext_small convnext_baseconvnext_largeconvnext_xlarge)r r)`)depthsprojection_dims default_size)rrr))rrri)r&r'r(i)tinysmallbaselargexlargea Instantiates the {name} architecture. References: - [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) (CVPR 2022) 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/). The `base`, `large`, and `xlarge` models were first pre-trained on the ImageNet-21k dataset and then fine-tuned on the ImageNet-1k dataset. The pre-trained parameters of the models were assembled from the [official repository](https://github.com/facebookresearch/ConvNeXt). To get a sense of how these parameters were converted to TF-Keras compatible parameters, please refer to [this repository](https://github.com/sayakpaul/keras-convnext-conversion). Note: Each TF-Keras Application expects a specific kind of input preprocessing. For ConvNeXt, preprocessing is included in the model using a `Normalization` layer. ConvNeXt models expect their inputs to be float or uint8 tensors of pixels with values in the [0-255] range. When calling the `summary()` method after instantiating a ConvNeXt model, prefer setting the `expand_nested` argument `summary()` to `True` to better investigate the instantiated model. 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-1k), 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. cB^\rSrSrSrU4SjrSSjrU4SjrSrU=r $)StochasticDepthaStochastic Depth module. It performs batch-wise dropping rather than sample-wise. In libraries like `timm`, it's similar to `DropPath` layers that drops residual paths sample-wise. References: - https://github.com/rwightman/pytorch-image-models Args: drop_path_rate (float): Probability of dropping paths. Should be within [0, 1]. Returns: Tensor either with the residual path dropped or kept. c 2>[TU]"S0UD6 XlgN)super__init__drop_path_rate)selfr6kwargs __class__s \/home/james-whalen/.local/lib/python3.13/site-packages/tf_keras/src/applications/convnext.pyr5StochasticDepth.__init__s "6",c4U(aSUR- n[R"U5S4S[[R"U55S- --nU[RR USS5-n[R "U5nX- U-$U$)Nr)r>)r6tfshapelenrandomuniformfloor)r7xr keep_probr@ random_tensors r:callStochasticDepth.calls D///IXXa[^%BHHQK0@10D(EEE% (9(9%A(FFMHH]3MM]2 2r<c^>[TU]5nURSUR05 U$)Nr6)r4 get_configupdater6r7configr9s r:rKStochasticDepth.get_configs-#% ')<)<=> r<)r6N) __name__ __module__ __qualname____firstlineno____doc__r5rHrK__static_attributes__ __classcell__r9s@r:r/r/s"-r<r/cD^\rSrSrSrU4SjrSrSrU4SjrSr U=r $) LayerScalezLayer scale module. References: - https://arxiv.org/abs/2103.17239 Args: init_values (float): Initial value for layer scale. Should be within [0, 1]. projection_dim (int): Projection dimensionality. Returns: Tensor multiplied to the scale. c >>[TU]"S0UD6 XlX lgr2)r4r5 init_valuesprojection_dim)r7r]r^r8r9s r:r5LayerScale.__init__s "6"&,r<cURSUR4[R"UR5SS9Ulg)NgammaT)namer@ initializer trainable) add_weightr^rConstantr]ra)r7 input_shapes r:buildLayerScale.builds=__&&($--d.>.>? %  r<cXR-$rP)ra)r7rEs r:rHLayerScale.calls::~r<ct>[TU]5nURURURS.5 U$)N)r]r^)r4rKrLr]r^rMs r:rKLayerScale.get_configs:#% #//"&"5"5   r<)rar]r^) rQrRrSrTrUr5rhrHrKrVrWrXs@r:rZrZs! -  r<rZcj^^^^Tc"S[[R"S55-mUUUU4SjnU$)aConvNeXt block. References: - https://arxiv.org/abs/2201.03545 - https://github.com/facebookresearch/ConvNeXt/blob/main/models/convnext.py Notes: In the original ConvNeXt implementation (linked above), the authors use `Dense` layers for pointwise convolutions for increased efficiency. Following that, this implementation also uses the same. Args: projection_dim (int): Number of filters for convolution layers. In the ConvNeXt paper, this is referred to as projection dimension. drop_path_rate (float): Probability of dropping paths. Should be within [0, 1]. layer_scale_init_value (float): Layer scale value. Should be a small float number. name: name to path to the keras layer. Returns: A function representing a ConvNeXtBlock block. prestemc>Un[R"TSSTTS-S9"U5n[R"STS-S9"U5n[R"ST-TS -S 9"U5n[R"S TS -S 9"U5n[R"TTS -S 9"U5nTb[ TTTS-S 9"U5nT(a[ TTS-S 9nO[R"STS-S 9nX"U5-$)Nsame_depthwise_conv)filters kernel_sizepaddinggroupsrbư> _layernormepsilonrb_pointwise_conv_1rbgelu_gelu_pointwise_conv_2 _layer_scale_stochastic_depthlinear _identity)rConv2DLayerNormalizationDense ActivationrZr/)inputsrElayerr6layer_scale_init_valuerbr^s r:applyConvNeXtBlock..applys  MM"!))     % %d 9L Ma P LL^+$9L2L Ma P   f4'> :1 = LLd5H.H I! L ! -&N* A #T,?%?E%%hTK5GHEa  r<strrget_uid)r^r6rrbrs```` r: ConvNeXtBlockrs34 |3wy9::!!< Lr<c^^Tc"S[[R"S55-mU4SjnU$)zsNormalizes inputs with ImageNet-1k mean and std. Args: name (str): Name prefix. Returns: A presemt function. rocN>[R"/SQ/SQTS-S9"U5nU$)N)g33333^@gR]@gRY@)g)~@gec}@gȷ@_prestem_normalization)meanvariancerb)r Normalization)rErbs r:rPreStem..apply<s7  8 00   r<r)rbrs` r:PreStemr0s. |3wy9::  Lr<c`^^^Tc[[R"S55mUUU4SjnU$)zImplementation of classification head of ConvNeXt. Args: num_classes: number of classes for Dense layer classifier_activation: activation function for the Dense layer name: name prefix Returns: Classification head function. headc>[R"TS-S9"U5n[R"STS-S9"U5n[R"TTTS-S9"U5nU$)N _head_gapr~rx_head_layernormrz _head_dense) activationrb)rGlobalAveragePooling2Drr)rEclassifier_activationrb num_classess r:rHead..applyYso  ) )tk/A B1 E  % %t&77   LL , %    r<r)rrrbrs``` r:HeadrKs) |7??6*+  Lr<c US;d9[RRRU5(d [ S5eUS:XaU(aU S:wa [ S5e[ R "U US[R"5UUS9n U c[R"U S9nO2[R"U 5(d[R"XS 9nOU nU b#[RRU 5S nOUnUnU(a:[R"5S :XaS OS nU US - nUS :Xa[US9"U5n[ R""[R$"US SSUS-S9[R&"SUS-S9/US-S9n/nUR)U5 S n[+U5Hn[ R""[R&"SUS-[-U5-S9[R$"UUS -SSUS-[-U5-S9/US-[-U5-S9nUR)U5 M [.R0"SU[3U55Vs/sHn[5U5PM nnS nSn[+U5HOnUU"W5n[+UU5H'nUUUU-UUSUSU3-S9"U5nM) UUU- nMQ U(a([ R8"X5 [;U U US9"W5nO]U S:Xa[R<"5"W5nO!U S :Xa[R>"5"W5n[R&"SS!9"W5n[@RB"UUUS"9nUS:XacU(aS#n[DUS nOS$n[DUS nUU-n[RFRIU[JU-S%US&9n URMU 5 U$UbURMU5 U$s snf)'a Instantiates ConvNeXt architecture given specific configuration. Args: depths: An iterable containing depths for each individual stages. projection_dims: An iterable containing output number of channels of each individual stages. drop_path_rate: Stochastic depth probability. If 0.0, then stochastic depth won't be used. layer_scale_init_value: Layer scale coefficient. If 0.0, layer scaling won't be used. 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. When `weights="imagenet"` this should be always set to True. But for other models (e.g., randomly initialized) users should set it to False and apply preprocessing to data accordingly. include_top: Boolean denoting whether to include classification head to the model. weights: one of `None` (random initialization), `"imagenet"` (pre-training on ImageNet-1k), 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 when using ImageNet. >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 1000 )r#min_size data_formatrequire_flattenweights)r@)tensorr@r channels_lastrr>r~r| _stem_conv)rustridesrbrx_stem_layernormrz_stem_downsampling_layernorm__downsampling_conv__downsampling_block__stage__block_)r^r6rrb)rrrbavgmax)r{)routputsrbz.h5z _notop.h5models) cache_subdir file_hash)'r?iogfileexists ValueErrorrobtain_input_shaperimage_data_formatrInputis_keras_tensorr layer_utilsget_source_inputsrr Sequentialrrappendrangernplinspacesumfloatrvalidate_activationrrGlobalMaxPooling2D training_libModelWEIGHTS_HASHES data_utilsget_fileBASE_WEIGHTS_PATH load_weights)!r!r"r6rr# model_nameinclude_preprocessing include_topr input_tensorrgpoolingclassesr img_inputrrE channel_axis num_channelsstemdownsample_layersnum_downsample_layersidownsample_layerdepth_drop_ratescurnum_convnext_blocksjmodel file_suffixr file_name weights_paths! r:ConvNeXtrhsED ) )RUU[[-?-?-H-H <  *D 1  !33!--/# KLL{3 &&|44 LLI$I""44\B1EA**,?AQ #3v;GGQaG C & ' a  #vay!A.q1/a8'=GA3gaS"99   A" vay(**+@J "7    e --/2A  ))+A.A  % %d 3A 6   faj IE* K&z215I%K&z215I, ''00   )! 1 <( L   7# LosO1z(keras.applications.convnext.ConvNeXtTinyzkeras.applications.ConvNeXtTinyc j[[SS[SSSS[SSUUUUUUUUUS9$)Nr)r!r"rrxr#r!r"r6rr#rrrrrrgrrrr MODEL_CONFIGS rrrrrrgrrrs r: ConvNeXtTinyr?Z V$X.%f-.?@#"6*>:3!3 r<z)keras.applications.convnext.ConvNeXtSmallz keras.applications.ConvNeXtSmallc j[[SS[SSSS[SSUUUUUUUUUS9$)Nr*r!r"rrxr#rrrs r: ConvNeXtSmallr`Z W%h/%g./@A#"7+N;3!3 r<z(keras.applications.convnext.ConvNeXtBasezkeras.applications.ConvNeXtBasec j[[SS[SSSS[SSUUUUUUUUUS9$)Nr+r!r"rrxr#rrrs r: ConvNeXtBaserrr<z)keras.applications.convnext.ConvNeXtLargez keras.applications.ConvNeXtLargec j[[SS[SSSS[SSUUUUUUUUUS9$)Nr,r!r"rrxr#rrrs r: ConvNeXtLargerrr<z*keras.applications.convnext.ConvNeXtXLargez!keras.applications.ConvNeXtXLargec j[[SS[SSSS[SSUUUUUUUUUS9$)Nr-r!r"rrxr#rrrs r:ConvNeXtXLargersZ X&x0%h/0AB#"8,^<3!3 r<rr~rrrrz,keras.applications.convnext.preprocess_inputcU$)aA placeholder method for backward compatibility. The preprocessing logic has been included in the convnext model implementation. Users are no longer required to call this method to normalize the input data. This method does nothing and only kept as a placeholder to align the API surface between old and new version of model. Args: x: A floating point `numpy.array` or a `tf.Tensor`. data_format: Optional data format of the image tensor/array. `None` means the global setting `tf.keras.backend.image_data_format()` is used (unless you changed it, it uses "channels_last"). Defaults to `None`. Returns: Unchanged `numpy.array` or `tf.Tensor`. r3)rErs r:preprocess_inputrs & Hr<z.keras.applications.convnext.decode_predictionsc*[R"XS9$)N)top)rdecode_predictions)predsrs r:rrs  , ,U <rs "!! %4*8:M #4.   .  0   0  1+ 8AH"fll"J''VKO;|6@   #Qn.%    # :/&     # :.%    # :/&     # :0' !   # :&,,.,A &--?-C %,,.,A &--?-C '..4D.E<= > *>?=@=,>>FFr<