U 9%eD@s&dZddlmZmZmZmZddlZddlmZddlm Z ddl m Z ddl m Z dd lmZdd lmZmZmZdd lmZd d lmZdgZdZGdddejZGdddejZGdddejZGdddejZGdddejZGdddeZ dZ!dZ"ede!Gdd d e Z#dS)!zs PyTorch UperNet model. Based on OpenMMLab's implementation, found in https://github.com/open-mmlab/mmsegmentation.)ListOptionalTupleUnionN)nn)CrossEntropyLoss) AutoBackbone)SemanticSegmenterOutput)PreTrainedModel)add_start_docstrings%add_start_docstrings_to_model_forwardreplace_return_docstrings) BackboneMixin) UperNetConfigzopenmmlab/upernet-convnext-tinyrc szeZdZdZd eeeeeeeffeeeeefefeeeeeeffddfdd Z e j e j d d d Z Z S) UperNetConvModulez A convolutional block that bundles conv/norm/activation layers. This block simplifies the usage of convolution layers, which are commonly used with a norm layer (e.g., BatchNorm) and activation layer (e.g., ReLU). rFrN) in_channels out_channels kernel_sizepaddingbiasdilationreturncs<ttj||||||d|_t||_t|_dS)N)rrrrrr) super__init__rConv2dconvZ BatchNorm2d batch_normZReLU activation)selfrrrrrr __class__k/var/www/html/Darija-Ai-API/env/lib/python3.8/site-packages/transformers/models/upernet/modeling_upernet.pyr.s  zUperNetConvModule.__init__inputrcCs"||}||}||}|SN)rrr)r r&outputr#r#r$forwardCs   zUperNetConvModule.forward)rFr)__name__ __module__ __qualname____doc__intrrstrboolrtorchTensorr) __classcell__r#r#r!r$r(s rcs<eZdZeeeddfdd ZejejdddZZS)UperNetPyramidPoolingBlockN) pool_scalerchannelsrcsLtt|t||ddg|_t|jD]\}}|t||q.dS)Nrr) rrrZAdaptiveAvgPool2drlayers enumerate add_moduler/)r r5rr6ilayerr!r#r$rLs   z#UperNetPyramidPoolingBlock.__init__r%cCs|}|jD] }||}q |Sr')r8)r r&Z hidden_stater<r#r#r$r)Us  z"UperNetPyramidPoolingBlock.forward) r*r+r,r.rr1r2r)r3r#r#r!r$r4Ks r4csNeZdZdZeedfeeeddfdd Zej e ej ddd Z Z S) UperNetPyramidPoolingModulea} Pyramid Pooling Module (PPM) used in PSPNet. Args: pool_scales (`Tuple[int]`): Pooling scales used in Pooling Pyramid Module. in_channels (`int`): Input channels. channels (`int`): Channels after modules, before conv_seg. align_corners (`bool`): align_corners argument of F.interpolate. .N) pool_scalesrr6 align_cornersrcsht||_||_||_||_g|_t|D]2\}}t|||d}|j || t ||q0dS)N)r5rr6) rrr>r?rr6blocksr9r4appendr:r/)r r>rr6r?r;r5blockr!r#r$rks  z$UperNetPyramidPoolingModule.__init__)xrcCsHg}|jD]8}||}tjj||ddd|jd}||q |S)Nbilinearsizemoder?)r@r functional interpolaterGr?rA)r rCZppm_outsppmZppm_outZupsampled_ppm_outr#r#r$r)ws  z#UperNetPyramidPoolingModule.forward) r*r+r,r-rr.r0rr1r2rr)r3r#r#r!r$r=\s" r=csLeZdZdZfddZddZddZdd Zej ej d d d Z Z S) UperNetHeadz Unified Perceptual Parsing for Scene Understanding. This head is the implementation of [UPerNet](https://arxiv.org/abs/1807.10221). cst||_|j|_||_|j|_d|_tj |j|j dd|_ t |j|jd|j|jd|_ t|jdt|j|j|jddd|_t|_t|_|jddD]@}t||jdd}t|j|jddd}|j||j|qtt|j|j|jddd|_dS)NFrr7)r?rrr)rrconfigr>rZ hidden_sizer6r?rr num_labels classifierr= psp_modulesrlen bottleneckZ ModuleList lateral_convs fpn_convsrAfpn_bottleneck)r rOrZl_convZfpn_convr!r#r$rs@    zUperNetHead.__init__cCs||jdSr'apply _init_weightsr r#r#r$ init_weightsszUperNetHead.init_weightscCs<t|tjr8|jjjd|jjd|jdk r8|jj dSNg)ZmeanZstd isinstancerrweightdataZnormal_rOZinitializer_rangerZzero_r moduler#r#r$rZs  zUperNetHead._init_weightscCs:|d}|g}|||tj|dd}||}|S)NrMrdim)extendrRr1catrT)r inputsrCZpsp_outsr(r#r#r$ psp_forwards  zUperNetHead.psp_forwardencoder_hidden_statesrcsfddtjDt}t|dddD]H}|djdd}|dtjj ||dj d|d<q@fd dt|dD}|dt|dddD]0}tjj |||djdddj d||<qt j |dd } |}|}|S) Ncsg|]\}}||qSr#r#).0r;Z lateral_conv)rkr#r$ sz'UperNetHead.forward..rrrMrDrErFcsg|]}j||qSr#)rV)rlr;)lateralsr r#r$rmsrd)r9rUrArirSrangeshaperrIrJr?r1rgrWrQ)r rkZused_backbone_levelsr;Z prev_shapeZfpn_outsr(r#)rkrnr r$r)s0   zUperNetHead.forward) r*r+r,r-rr\rZrir1r2r)r3r#r#r!r$rLs  ' rLcsbeZdZdZdeeeeeeeffddfdd Zd d Zd d Z e j e j d ddZ Z S)UperNetFCNHeada Fully Convolution Networks for Semantic Segmentation. This head is the implementation of [FCNNet](https://arxiv.org/abs/1411.4038>). Args: config: Configuration. in_channels (int): Number of input channels. kernel_size (int): The kernel size for convs in the head. Default: 3. dilation (int): The dilation rate for convs in the head. Default: 1. rDrrN)in_indexrrrc st||_|j|_|j|_|j|_|j |_ ||_ |d|}g}| t |j|j|||dt|jdD] }| t |j|j|||dqp|jdkrt|_n tj||_|j rt |j|j|j||dd|_tj|j|jdd|_dS)NrD)rrrrrrNr7)rrrOZauxiliary_in_channelsrZauxiliary_channelsr6Zauxiliary_num_convsZ num_convsZauxiliary_concat_input concat_inputrrrArrorZIdentityconvsZ Sequentialconv_catrrPrQ)r rOrrrrZ conv_paddingrtr;r!r#r$rsN      zUperNetFCNHead.__init__cCs||jdSr'rXr[r#r#r$r\szUperNetFCNHead.init_weightscCs<t|tjr8|jjjd|jjd|jdk r8|jj dSr]r^rbr#r#r$rZs  zUperNetFCNHead._init_weightsrjcCs@||j}||}|jr2|tj||gdd}||}|S)Nrrd)rrrtrsrur1rgrQ)r rk hidden_statesr(r#r#r$r)s    zUperNetFCNHead.forward)rDrr)r*r+r,r-r.rrrr\rZr1r2r)r3r#r#r!r$rqs$rqc@s6eZdZdZeZdZdZddZddZ d d d Z d S) UperNetPreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. pixel_valuesTcCs6t|tr2|j|j|jdk r2|jdSr')r_rwbackboner\ decode_headauxiliary_headrbr#r#r$rZ0s     z$UperNetPreTrainedModel._init_weightscCs,|j|j|jdk r(|jdS)zInitialize the weightsN)ryr\rzr{r[r#r#r$r\7s   z#UperNetPreTrainedModel.init_weightsFcCst|tr||_dSr')r_rZgradient_checkpointing)r rcvaluer#r#r$_set_gradient_checkpointing>s z2UperNetPreTrainedModel._set_gradient_checkpointingN)F) r*r+r,r-r config_classZmain_input_nameZsupports_gradient_checkpointingrZr\r}r#r#r#r$rw&srwaI Parameters: This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. config ([`UperNetConfig`]): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights. ax Args: pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using [`AutoImageProcessor`]. See [`SegformerImageProcessor.__call__`] for details. output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers in case the backbone has them. See `attentions` under returned tensors for more detail. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers of the backbone. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. zMUperNet framework leveraging any vision backbone e.g. for ADE20k, CityScapes.c speZdZfddZeedeee dd e e j e e e e e e j e e eeefdddZZS) UperNetForSemanticSegmentationcsLt|t|j|_t||jjd|_|j r:t |nd|_ | dS)N)r) rrr from_configZbackbone_configryrLr6rzZuse_auxiliary_headrqr{Z post_init)r rOr!r#r$rbs  z'UperNetForSemanticSegmentation.__init__zbatch_size, sequence_length) output_typer~N)rxoutput_attentionsoutput_hidden_stateslabels return_dictrcCsp|dk r |n|jj}|dk r |n|jj}|dk r4|n|jj}|jj|||d}|j}||}tj j ||j ddddd}d} |j dk r| |} tj j | |j ddddd} d} |dk r|jj dkrtdn 1`, a classification loss is computed (Cross-Entropy). Returns: Examples: ```python >>> from transformers import AutoImageProcessor, UperNetForSemanticSegmentation >>> from PIL import Image >>> from huggingface_hub import hf_hub_download >>> image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-tiny") >>> model = UperNetForSemanticSegmentation.from_pretrained("openmmlab/upernet-convnext-tiny") >>> filepath = hf_hub_download( ... repo_id="hf-internal-testing/fixtures_ade20k", filename="ADE_val_00000001.jpg", repo_type="dataset" ... ) >>> image = Image.open(filepath).convert("RGB") >>> inputs = image_processor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits # shape (batch_size, num_labels, height, width) >>> list(logits.shape) [1, 150, 512, 512] ```N)rrrDrEFrFrz/The number of labels should be greater than one)Z ignore_index)losslogitsrv attentions)rOZuse_return_dictrrryZforward_with_filtered_kwargsZ feature_mapsrzrrIrJrpr{rP ValueErrorrZloss_ignore_indexZauxiliary_loss_weightr rvr)r rxrrrroutputsfeaturesrZauxiliary_logitsrZloss_fctZauxiliary_lossr(r#r#r$r)nsR(          z&UperNetForSemanticSegmentation.forward)NNNNN)r*r+r,rr UPERNET_INPUTS_DOCSTRINGformatrr _CONFIG_FOR_DOCrr1r2r0rtupler)r3r#r#r!r$r]s    r)$r-typingrrrrr1rZtorch.nnrr Zmodeling_outputsr Zmodeling_utilsr utilsr r rZutils.backbone_utilsrZconfiguration_upernetrZ%UPERNET_PRETRAINED_MODEL_ARCHIVE_LISTrModulerr4r=rLrqrwZUPERNET_START_DOCSTRINGrrr#r#r#r$s4       #&]G