^jddlZddlmZmZddlmZddlZddlZddl m cm Z ddlm Z m Z ddlmZddlmZddlmZmZdd lmZdd lmZdd lmZmZmZmZmZdd l m!Z!m"Z"dd l#m$Z$ddl%m&Z&ddl'm(Z(m)Z)ddl*m+Z+m,Z,m-Z-ddl.m/Z/ddl0m1Z1e)deGddeZ2e)deGddeZ3e)deGddeZ4e)deGdd eZ5e)d!eGd"d#eZ6Gd$d%e jnZ8e&d&'d(e9d)e9d*ejtd+ejvd,ejf d-Z< dd.ejd/e=dzd0e=dzd1e=dzd,ejf d2Z>Gd3d4e jnZ?ed5Gd6d7e jnZ@d8ZA dd9e jnd:ejd;ejde=e9zd?e=dzd@e=dzd,eBejejffdAZCdBejdCejdDejdEejd,eBejejff dFZDdGejdHe9d,ejfdIZEGdJdKe jnZFGdLdMe jnZGGdNdOe jnZHGdPdQe jnZIGdRdSe jnZJGdTdUeZKGdVdWe jnZLGdXdYe jnZMGdZd[e jnZNGd\d]e jnZOGd^d_e jnZPe)Gd`dae"ZQGdbdceQZRe)GdddeeQZSe)GdfdgeeQZTe)dhiGdjdkeQZUddlZVe)dmdnoGdpdqeQZWe)drdsoGdtdueQZXe)dvdwoGdxdyeQZYe)dzd{oGd|d}eQZZgd~Z[y)N)CallableIterable) dataclass)Tensornn)initialization)ACT2FN) BackboneMixinfilter_output_hidden_states)use_kernel_forward_from_hub)GradientCheckpointingLayer)BackboneOutputBaseModelOutputBaseModelOutputWithPooling ModelOutputSemanticSegmenterOutput)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)#compile_compatible_method_lru_cache)TransformersKwargsauto_docstring)can_return_tuplemaybe_autocastmerge_with_config_defaults)capture_outputs)Sapiens2Configz Output type of [`Sapiens2Backbone`], extending [`BackboneOutput`] with optional CLS tokens from each selected feature stage (used when `config.return_class_token=True`). ) custom_introc@eZdZUdZdZeejdzed<y)Sapiens2BackboneOutputz cls_tokens (`tuple(torch.FloatTensor)`, *optional*): CLS token from each selected feature stage, each of shape `(batch_size, hidden_size)`. Only present when `config.return_class_token=True`. N cls_tokens) __name__ __module__ __qualname____doc__r#tupletorch FloatTensor__annotations__y/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/sapiens2/modeling_sapiens2.pyr"r"1s# 37Je''(4/6r-r"z6 Class for outputs of pose estimation models. ceZdZUdZdZej dzed<dZej dzed<dZ e ej dfdzed<dZ e ej dfdzed<y)Sapiens2PoseEstimatorOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Pose estimation loss. heatmaps (`torch.FloatTensor` of shape `(batch_size, num_keypoints, height, width)`): Heatmaps as predicted by the model. hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called feature maps) of the model at the output of each stage. Nlossheatmaps. hidden_states attentions) r$r%r&r'r1r)r*r+r2r3r(r4r,r-r.r0r0Esq &*D%  d "))-He$&-:>M5**C/047>7;Je'',-4;r-r0z8 Class for outputs of normal estimation models. ceZdZUdZdZej dzed<dZej dzed<dZ e ej dfdzed<dZ e ej dfdzed<y)Sapiens2NormalEstimatorOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Normal estimation loss. normals (`torch.FloatTensor` of shape `(batch_size, num_labels, height, width)`): Raw normal map predictions as output by the model (unnormalized). hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs. attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`): Tuple of `torch.FloatTensor` (one per layer) of shape `(batch_size, num_heads, sequence_length, sequence_length)`. Attentions weights after the attention softmax. Nr1normals.r3r4) r$r%r&r'r1r)r*r+r7r3r(r4r,r-r.r6r6]sq &*D%  d ")(,GU   %,:>M5**C/047>7;Je'',-4;r-r6z: Class for outputs of pointmap estimation models. ceZdZUdZdZej dzed<dZej dzed<dZ ej dzed<dZ e ej dfdzed<dZ e ej dfdzed<y) Sapiens2PointmapEstimatorOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Pointmap estimation loss. pointmaps (`torch.FloatTensor` of shape `(batch_size, 3, height, width)`): Per-pixel 3D XYZ coordinate predictions in canonical camera space. scales (`torch.FloatTensor` of shape `(batch_size, 1)`, *optional*): Canonical focal length / actual focal length ratio. `None` when no scale branch is configured. hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs. attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`): Tuple of `torch.FloatTensor` (one per layer) of shape `(batch_size, num_heads, sequence_length, sequence_length)`. Attentions weights after the attention softmax. Nr1 pointmapsscales.r3r4) r$r%r&r'r1r)r*r+r:r;r3r(r4r,r-r.r9r9xs &*D%  d ")*.Iu  4'.'+FE   $+:>M5**C/047>7;Je'',-4;r-r9z4 Class for outputs of image matting models. ceZdZUdZdZej dzed<dZej dzed<dZ e ej dzed<dZ e ej dzed<dZ ej dzed<y)Sapiens2ImageMattingOutputaN loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Loss. alphas (`torch.FloatTensor` of shape `(batch_size, 1, height, width)`): Estimated alpha values. hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called feature maps) of the model at the output of each stage. foregrounds (`torch.FloatTensor` of shape `(batch_size, 3, height, width)`): Pre-multiplied RGB foreground predictions in `[0, 1]` (sigmoid-activated). Nr1alphasr3r4 foregrounds) r$r%r&r'r1r)r*r+r>r3r(r4r?r,r-r.r=r=s| &*D%  d ")'+FE   $+59M5**+d2926Je''(4/6,0K""T)0r-r=ceZdZdZdeffd Zd dejdejdzdejfdZxZ S) Sapiens2EmbeddingszM Construct the CLS token, mask token, position and patch embeddings. configcBt|||_tjt j dd|j|_|jr4tjt jdd|jnd|_ tjt jd|j|j|_tj|j |j|j"|j"|_y)Nr) kernel_sizestride)super__init__rBr Parameterr)randn hidden_size cls_tokenuse_mask_tokenzeros mask_tokenemptynum_register_tokensregister_tokensConv2d num_channels patch_sizepatch_embeddingsselfrB __class__s r.rGzSapiens2Embeddings.__init__s  ekk!Q8J8J&KLQWQfQf",,u{{1a9K9K'LMlp!||EKK6;U;UW]WiWi,jk "   !3!3ARAR[a[l[l! r-N pixel_valuesbool_masked_posreturnc\||j td|jd}|jjj }|j|j |}|jdjdd}|K|jj |j }tj|jd||}|jj|dd}|jj|dd}tj|||gd} | S)Nz:bool_masked_pos requires use_mask_token=True in the configrdtyperdim)rN ValueErrorshaperUweightr^toflatten transposer)where unsqueezerKexpandrQcat) rWrYrZ batch_size target_dtyperUrNrKrQ embeddingss r.forwardzSapiens2Embeddings.forwards  &4??+BYZ Z!''* ,,3399  00|1TU+33A6@@AF  &++,<,B,BCJ${{?+D+DR+H*Vfg NN))*b"= ..55j"bIYY ?ZZF^^Aq)F[[^F))F#C))F#C+ $.""vvEv"CFFF$7771 $ $s C7GG ) r$r%r&r)rr+rrGr(rprrrss@r.rrsBll5~5" 8ELL 8U5<<;U5V 8r-rRMSNormcheZdZddeddffd Zdej dej fdZdZxZ S) Sapiens2RMSNormepsr[Nct|tjt j ||_||_y)z> Sapiens2RMSNorm is equivalent to T5LayerNorm N)rFrGrrHr)onesrevariance_epsilon)rWrJrrXs r.rGzSapiens2RMSNorm.__init__Js1 ll5::k#:; #r-r3c"|j}|jtj}|j dj dd}|tj ||jzz}|j|j|zS)Nr_r`T)keepdim) r^rfr)rpowmeanrsqrtrre)rWr3 input_dtypevariances r.rpzSapiens2RMSNorm.forwardRsy#)) %((7  $$Q',,R,>% Ht?T?T4T(UU {{]--k:::r-c^t|jjd|jS)Nz, eps=)r(rerdrrWs r. extra_reprzSapiens2RMSNorm.extra_reprYs* ))*+6$2G2G1HIIr-)gư>) r$r%r&floatrGr)rrprrrrss@r.rrHs7$$$$;U\\;ell;Jr-rc|dd|jddzf}|d|jddzdf}tj| |fdS)z*Rotates half the hidden dims of the input..Nr`r_ra)rdr)rl)xx1x2s r. rotate_halfr]sZ 3"!''"+"" " #B 3 q " " #B 99rc2YB ''r-modulequerykeyvalueattention_maskdropoutscalingsoftcapc |||jdz}t||j} t||j} tj|| j dd|z} || |z } tj | } | |z} || |z} tjj| dtjj|j} tjj| ||j} tj| | } | j ddj} | | fS)Nr_rr`)rbr^)prr)r repeat_kvnum_key_value_groupsr)matmulrhtanhr functionalsoftmaxrrfr^rr contiguous) rrrrrrrrkwargs key_states value_states attn_weights attn_outputs r.eager_attention_forwardrds//4'3 ; ;