3j@BSSKrSSKJrJr SSKJr SSKrSSKrSSKJ r J r SSK J r SSKJr SSKJrJr SS KJr SS KJr SS KJrJrJrJrJr SS KJrJr SS K J!r! SSK"J#r# SSK$J%r%J&r& SSK'J(r(J)r)J*r* SSK+J,r, SSK-J.r. \&"SS9\"SS\555r/\&"SS9\"SS\555r0\&"SS9\"SS\555r1\&"SS9\"SS \555r2\&"S!S9\"S"S#\555r3"S$S%\ Rh5r5\#"S&S'9S(\6S)\6S*\RnS+\RpS,\R4 S-j5r9SS.\RS/\:S-S0\:S-S1\:S-S,\R4 S2jjr;"S3S4\ Rh5r<\"S55"S6S7\ Rh55r=S8r>SS9\ RhS:\RS;\RS<\RS=\RS-S>\:\6-S?\:S-S@\:S-S,\?\R\R44SAjjr@SB\RSC\RSD\RSE\RS,\?\R\R44 SFjrASG\RSH\6S,\R4SIjrB"SJSK\ Rh5rC"SLSM\ Rh5rD"SNSO\ Rh5rE"SPSQ\ Rh5rF"SRSS\ Rh5rG"STSU\5rH"SVSW\ Rh5rI"SXSY\ Rh5rJ"SZS[\ Rh5rK"S\S]\ Rh5rL"S^S_\ Rh5rM\&"S`Sa\55rN"SbSc\N5rO\&"SdSe\N55rP\&"SfSg\\N55rQ\&"ShSi9"SjSk\N55rRSSljrS\&"SmSnSo9"SpSq\N55rT\&"SrSsSo9"StSu\N55rU\&"SvSwSo9"SxSy\N55rV\&"SzS{So9"S|S}\N55rW/S~QrXg)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_introcH\rSrSr%SrSr\\RS-\ S'Sr g)Sapiens2BackboneOutput0z 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____firstlineno____doc__r$tupletorch FloatTensor__annotations____static_attributes__r%h/home/wildlama/miniconda3/lib/python3.13/site-packages/transformers/models/sapiens2/modeling_sapiens2.pyr"r"0s# 37Je''(4/6r0r"z6 Class for outputs of pose estimation models. c\rSrSr%SrSr\RS-\S'Sr \RS-\S'Sr \ \RS4S-\S'Sr \ \RS4S-\S'S r g) Sapiens2PoseEstimatorOutputDa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Pose estimation loss. heatmaps (`torch.FloatTensor` of shape `(batch_size, num_keypoints, height, width)`): Heatmaps as predicted by the model. hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called feature maps) of the model at the output of each stage. Nlossheatmaps. hidden_states attentionsr%)r&r'r(r)r*r5r,r-r.r6r7r+r8r/r%r0r1r3r3Dsq &*D%  d "))-He$&-:>M5**C/047>7;Je'',-4;r0r3z8 Class for outputs of normal estimation models. c\rSrSr%SrSr\RS-\S'Sr \RS-\S'Sr \ \RS4S-\S'Sr \ \RS4S-\S'S r g) Sapiens2NormalEstimatorOutput\a 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. Nr5normals.r7r8r%)r&r'r(r)r*r5r,r-r.r<r7r+r8r/r%r0r1r:r:\sq &*D%  d ")(,GU   %,:>M5**C/047>7;Je'',-4;r0r:z: Class for outputs of pointmap estimation models. c\rSrSr%SrSr\RS-\S'Sr \RS-\S'Sr \RS-\S'Sr \ \RS4S-\S'Sr \ \RS4S-\S 'S rg) Sapiens2PointmapEstimatorOutputwa 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. Nr5 pointmapsscales.r7r8r%)r&r'r(r)r*r5r,r-r.r@rAr7r+r8r/r%r0r1r>r>ws &*D%  d ")*.Iu  4'.'+FE   $+:>M5**C/047>7;Je'',-4;r0r>z4 Class for outputs of image matting models. c\rSrSr%SrSr\RS-\S'Sr \RS-\S'Sr \ \RS-\S'Sr \ \RS-\S'Sr \RS-\S'S rg) Sapiens2ImageMattingOutputa" 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). Nr5alphasr7r8 foregroundsr%)r&r'r(r)r*r5r,r-r.rEr7r+r8rFr/r%r0r1rCrCs| &*D%  d ")'+FE   $+59M5**+d2926Je''(4/6,0K""T)0r0rCc^\rSrSrSrS\4U4SjjrS S\RS\RS-S\R4S jjr S r U=r $) Sapiens2EmbeddingszE Construct the CLS token, mask token, position and patch embeddings. configcR>[TU]5 Xl[R"[ R "SSUR55UlUR(a6[R"[ R"SSUR55OSUl [R"[ R"SURUR55Ul[R"UR URUR"UR"S9Ulg)Nr) kernel_sizestride)super__init__rJr Parameterr,randn hidden_size cls_tokenuse_mask_tokenzeros mask_tokenemptynum_register_tokensregister_tokensConv2d num_channels patch_sizepatch_embeddingsselfrJ __class__s r1rOSapiens2Embeddings.__init__s  ekk!Q8J8J&KLQWQfQf",,u{{1a9K9K'LMlp!||EKK6;U;UW]WiWi,jk "   !3!3ARAR[a[l[l! r0N pixel_valuesbool_masked_posreturncVUbURc [S5eURSnURRR nURUR US95nURS5RSS5nUbKURR UR 5n[R"URS5Xe5nURRUSS5nURRUSS5n[R"XxU/SS9n U $)Nz:bool_masked_pos requires use_mask_token=True in the configrdtyperdim)rV ValueErrorshaper]weightrgtoflatten transposer,where unsqueezerSexpandrYcat) r_rbrc batch_size target_dtyper]rVrSrY embeddingss r1forwardSapiens2Embeddings.forwards  &4??+BYZ Z!''* ,,3399  00|1TU+33A6@@AF  &++,<,B,BCJ${{?+D+DR+H*g NN))*b"= ..55j"bIYY [T U]5 XlURUlUR UR -UlSUR[R"SSSUR- [RS9-- nURSUSS9 URn[U[5(aUOX34upEURn[U[ 5(aUOUSn[U[ 5(aUOUSnXG-UlXX-Ulg)NrrrfrF) persistent)rNrOrJ rope_thetabaserRnum_attention_headshead_dimr,rfloat32register_buffer image_size isinstancerr\intrr) r_rJrrimage_himage_wr\ patch_size_h patch_size_wr`s r1rO&Sapiens2RopePositionEmbedding.__init__s  %% **f.H.HH tyyELLAq4==7HPUP]P]$^^^ ZeD&& )3J)I)I:PZOg&& %/ C%@%@zjQRm %/ C%@%@zjQRm $4$4r0rbrdcURu p#nX0RR-nX@RR-nURn[ UR [ 5(aUR S:wa UR OSn[USS9 [XV[RUS9n UR(aH[U URRURRURRS9n S[ R"-U SS2SS2S4-UR$SSSS24-n U R'SS5n U R)S5n [R*"U 5n [R,"U 5n SSS5 UR.n W R1U S 9W R1U S 94$!,(df  N8=f) NmpscpuF) device_typeenabledr)rrrrhrrf)rmrJr\rrtypestrrrr,rtrainingrpos_embed_shiftpos_embed_jitterpos_embed_rescalemathpirrptilecossinrgro)r_rb_heightwidthrrrr patch_coordsanglesrrrgs r1ry%Sapiens2RopePositionEmbedding.forward$su*001e++"8"88 !7!77 $$%/ S%A%AfkkUZFZfkk`e  U C:EMM&L}}A ++55;;77 KK99   [<1d #;;dmmDRVXYM>ZZF^^Aq)F[[^F))F#C))F#C+D.""vvEv"CFFF$7771D Cs C7F== G )rrJrrr) r&r'r(r)r,rr.rrOr+ryr/r|r}s@r1rrsGll5~5" 8ELL 8U5<<;U5V 8 8r0rRMSNormcx^\rSrSrS S\SS4U4SjjjrS\RS\R4SjrSr S r U=r $) Sapiens2RMSNormiGepsrdNc>[TU]5 [R"[R "U55UlX lg)z. Sapiens2RMSNorm is equivalent to T5LayerNorm N)rNrOrrPr,onesrnvariance_epsilon)r_rRrr`s r1rOSapiens2RMSNorm.__init__Is/ ll5::k#:; #r0r7cURnUR[R5nUR S5R SSS9nU[R "X0R-5-nURURU5-$)NrhriT)keepdim) rgror,rpowmeanrsqrtrrn)r_r7 input_dtypevariances r1rySapiens2RMSNorm.forwardQsw#)) %((7  $$Q',,R,>% H?T?T4T(UU {{]--k:::r0c^[URR5SUR3$)Nz, eps=)r+rnrmrr_s r1 extra_reprSapiens2RMSNorm.extra_reprXs* ))*+6$2G2G1HIIr0)rrn)gư>) r&r'r(r)floatrOr,rryrr/r|r}s@r1rrGsB$$$$$;U\\;ell;JJr0rcUSSURSS-24nUSURSS-S24n[R"U*U4SS9$)z*Rotates half the hidden dims of the input..Nrirhrj)rmr,ru)xx1x2s r1 rotate_halfr\sZ 3"!''"+"" " #B 3 q " " #B 99rc2YB ''r0modulequerykeyvalueattention_maskdropoutscalingsoftcapc jUcURS-n[X R5n [X0R5n [R"XR SS55U-n UbX- n [R "U 5n X-n UbX-n [RRU S[RS9RUR5n [RRXURS9n [R"X5n U R SS5R5n X4$)Nrhrri)rkrg)prr)r repeat_kvnum_key_value_groupsr,matmulrqtanhr functionalsoftmaxrrorgrr contiguous) rrrrrrrrkwargs key_states value_states attn_weights attn_outputs r1eager_attention_forwardrcs //4'3 ; ;