9im<SSKJrJr SSKrSSKrSSKrSSKJr SSK Js J r SSK J r SSKJrJrJr "SS\R$5rSSjr"SS \R$5r"S S \5r"S S \\S9r"SS\5r"SS\R$5rg))ABCMetaabstractmethodN) ConvModule) BaseModule auto_fp16 force_fp32c.^\rSrSrU4SjrSrSrU=r$)LinearClassifierc`>[[U] 5 [R"X5UlgN)superr __init__nnLinear classifier)self in_channels num_classes __class__s k/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/modelscope/models/cv/vision_middleware/head.pyrLinearClassifier.__init__s! .0))K=c*URUS5$)Nr)rxs rforwardLinearClassifier.forwardsqu%%rr__name__ __module__ __qualname____firstlineno__rr__static_attributes__ __classcell__rs@rr r s>&&rr c2[R"XX#U5$r )F interpolate)inputsize scale_factormode align_cornerswarnings rresizer1s ==l- HHrc:^\rSrSrSU4SjjrSrSrU=r$)Upsample%c>[[U] 5 Xl[ U[ 5(a[ SU55UlOU(a [U5OSUlX0lX@l g)Nc38# UHn[U5v M g7fr )float).0factors r $Upsample.__init__../s%O,eFmm,s) rr3rr, isinstancetupler-r7r.r/)rr,r-r.r/rs rrUpsample.__init__'sS h&( lE * * %%O,%O OD 7Cl 3D  *rcUR(d5URSSVs/sHn[X R-5PM nnO URn[ XSUR UR 5$s snf)N)r,shapeintr-r1r.r/)rrtr,s rrUpsample.forward5sayy89 E 1C---. DED99DatTYY0B0BCCFs!A5)r/r.r-r,)NNnearestNr r's@rr3r3%s""# +DDrr3ct^\rSrSrSrSSSSSSSSS\"SS9\"S S S S 94 U4S jjr\"5S5rSr U=r $)FPN=aJ Feature Pyramid Network. This neck is the implementation of `Feature Pyramid Networks for Object Detection `_. Args: in_channels (list[int]): Number of input channels per scale. out_channels (int): Number of output channels (used at each scale). num_outs (int): Number of output scales. start_level (int): Index of the start input backbone level used to build the feature pyramid. Default: 0. end_level (int): Index of the end input backbone level (exclusive) to build the feature pyramid. Default: -1, which means the last level. add_extra_convs (bool | str): If bool, it decides whether to add conv layers on top of the original feature maps. Default to False. If True, its actual mode is specified by `extra_convs_on_inputs`. If str, it specifies the source feature map of the extra convs. Only the following options are allowed - 'on_input': Last feat map of neck inputs (i.e. backbone feature). - 'on_lateral': Last feature map after lateral convs. - 'on_output': The last output feature map after fpn convs. extra_convs_on_inputs (bool, deprecated): Whether to apply extra convs on the original feature from the backbone. If True, it is equivalent to `add_extra_convs='on_input'`. If False, it is equivalent to set `add_extra_convs='on_output'`. Default to True. relu_before_extra_convs (bool): Whether to apply relu before the extra conv. Default: False. no_norm_on_lateral (bool): Whether to apply norm on lateral. Default: False. conv_cfg (dict): Config dict for convolution layer. Default: None. norm_cfg (dict): Config dict for normalization layer. Default: None. act_cfg (dict): Config dict for activation layer in ConvModule. Default: None. upsample_cfg (dict): Config dict for interpolate layer. Default: dict(mode='nearest'). init_cfg (dict or list[dict], optional): Initialization config dict. Example: >>> import torch >>> in_channels = [2, 3, 5, 7] >>> scales = [340, 170, 84, 43] >>> inputs = [torch.rand(1, c, s, s) ... for c, s in zip(in_channels, scales)] >>> self = FPN(in_channels, 11, len(in_channels)).eval() >>> outputs = self.forward(inputs) >>> for i in range(len(outputs)): ... print(f'outputs[{i}].shape = {outputs[i].shape}') outputs[0].shape = torch.Size([1, 11, 340, 340]) outputs[1].shape = torch.Size([1, 11, 170, 170]) outputs[2].shape = torch.Size([1, 11, 84, 84]) outputs[3].shape = torch.Size([1, 11, 43, 43]) rrFNrE)r.XavierConv2duniform)typelayer distributionc 4>[[U] U5 [U[5(deXlX l[U5UlX0l Xl Xl SUl U R5UlUS:Xa&URUlX0RU- :deO!XPlU[U5::deX5U- :XdeX@lXPlX`l[U[&[(45(de[U[&5(a US;deOU(aU(aSUlOSUl[*R,"5Ul[*R,"5Ul[3UR UR5Hmn[5XUSU UR(dU OSU SS9n[5UUSSU U U SS 9nUR.R7U5 UR0R7U5 Mo X0R- UR -nUR$(a{US:at[3U5HdnUS :Xa-UR$S:XaUR URS- nOUn[5UUSS SU U U SS 9 nUR0R7U5 Mf ggg) NFr)on_input on_lateral on_outputrPrR)conv_cfgnorm_cfgact_cfginplace)paddingrTrUrVrWr)striderYrTrUrVrW)rrGrr<listr out_channelslennum_insnum_outsrelu_before_extra_convsno_norm_on_lateral fp16_enabledcopy upsample_cfgbackbone_end_level start_level end_leveladd_extra_convsstrboolr ModuleList lateral_convs fpn_convsrangerappend)rrr]r`rgrhriextra_convs_on_inputsrarbrTrUrVreinit_cfgil_convfpn_conv extra_levelsextra_fpn_convrs rr FPN.__init__psx c4!(++t,,,,&(;'  '>$"4!(--/ ?&*llD #||k99 99'0 #K 00 00;66 66&"./C;7777 os + +"&MM MM $(2$'2$]]_t'')@)@AA!)-)@)@dF"!!H    % %f - NN ! !( +)B. "9"99D[[U] U5 URXU 5 X lX`lXplXlXlXl Xl Xl UcUS:Xa[RS5 UnXC:waUS:wa[SUSU35eUS:XaUcSn[RS5 X0lX@lXPl[$R&"X R SS9UlUS :a[$R*"U5UlOSUlS Ulg) NrZzFor binary segmentation, we suggest using`out_channels = 1` to define the outputchannels of segmentor, and use `threshold`to convert seg_logist into a predictionapplying a thresholdrSzout_channels should be equal to num_classes,except binary segmentation set out_channels == 1 andnum_classes == 2, but got out_channels=zand num_classes=g333333?z7threshold is not defined for binary, and defaultsto 0.3) kernel_sizerF)rrr _init_inputschannelsrrTrUrVrrr/warningswarn ValueErrorrr]rrrJr Dropout2ddropoutrc)rrrrr]rrrTrUrVrrrrrr/rrrs rrBaseDecodeHead.__init__/s, nd,X6 +A *    (*  a 56 'L  &<1+<::F";-12 2 1 !2I MM# $&(" (,=,=1M 1 << 6DLDL!rcXSURSURSUR3nU$)z Extra repr.zinput_transform=z, ignore_index=z, align_corners=)rrr/)rss r extra_reprBaseDecodeHead.extra_reprps@t3345 --./!//0 2rcUbUS;deX0lX lUbr[U[[45(de[U[[45(de[ U5[ U5:XdeUS:Xa[ U5UlgXlg[U[5(de[U[5(deXlg)aMCheck and initialize input transforms. The in_channels, in_index and input_transform must match. Specifically, when input_transform is None, only single feature map will be selected. So in_channels and in_index must be of type int. When input_transform Args: in_channels (int|Sequence[int]): Input channels. in_index (int|Sequence[int]): Input feature index. input_transform (str|None): Transformation type of input features. Options: 'resize_concat', 'multiple_select', None. 'resize_concat': Multiple feature maps will be resize to the same size as first one and than concat together. Usually used in FCN head of HRNet. 'multiple_select': Multiple feature maps will be bundle into a list and passed into decode head. None: Only one select feature map is allowed. N) resize_concatmultiple_selectr) rrr<r\r=r^sumrrB)rrrrs rrBaseDecodeHead._init_inputsws&  &"&JJ JJ.  &kD%=99 99hu 66 66{#s8}4 44/1#&{#3 #. k3// //h,, ,,* rc URS:XakURVs/sHo!UPM nnUVs/sH)n[UUSRSSSURS9PM+ nn[ R "USS9nU$URS :Xa URVs/sHo!UPM nnU$XRnU$s snfs snfs snf) zTransform inputs for decoder. Args: inputs (list[Tensor]): List of multi-level img features. Returns: Tensor: The transformed inputs rrrZNbilinear)r+r,r.r/rSdimr)rrr1rAr/torchcat)rr~rsrupsampled_inputss r_transform_inputs BaseDecodeHead._transform_inputss   ? 2)-7AQiF7 @F   @F! ,#"&"4"4 6@F  YY/Q7F   ! !%6 6)-7AQiF7 MM*F 8 8sB<0CCcg)z Placeholder of forward function.N)rr~s rrBaseDecodeHead.forwards rc8U"U5nURXS5nU$)aForward function for training. Args: inputs (list[Tensor]): List of multi-level img features. img_metas (list[dict]): List of image info dict where each dict has: 'img_shape', 'scale_factor', 'flip', and may also contain 'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'. For details on the values of these keys see `mmseg/datasets/pipelines/formatting.py:Collect`. gt_semantic_seg (Tensor): Semantic segmentation masks used if the architecture supports semantic segmentation task. train_cfg (dict): The training config. Returns: dict[str, Tensor]: a dictionary of loss components )losses)rr~ img_metasgt_semantic_seg train_cfg seg_logitsrs r forward_trainBaseDecodeHead.forward_trains &\ Z9 rc$URU5$)aForward function for testing. Args: inputs (list[Tensor]): List of multi-level img features. img_metas (list[dict]): List of image info dict where each dict has: 'img_shape', 'scale_factor', 'flip', and may also contain 'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'. For details on the values of these keys see `mmseg/datasets/pipelines/formatting.py:Collect`. test_cfg (dict): The testing config. Returns: Tensor: Output segmentation map. )r)rr~rtest_cfgs r forward_testBaseDecodeHead.forward_tests||F##rcdURbURU5nURU5nU$)zClassify each pixel.)rr)rfeatoutputs rcls_segBaseDecodeHead.cls_segs- << #<<%Dt$ r)rVr/rrTrrrrcrrrrrUrr]r)r!r"r#r$rrrrrrrrrrrrr%r&r's@rrrs'\#"6*!%!,!&!$&"$"t7LN)?"?"B"+H2[  & $rr) metaclassc2^\rSrSrSrU4SjrSrSrU=r$)FPNHeadia3Panoptic Feature Pyramid Networks. This head is the implementation of `Semantic FPN `_. Args: feature_strides (tuple[int]): The strides for input feature maps. stack_lateral. All strides suppose to be power of 2. The first one is of largest resolution. c R>[[U] "S SS0UD6 [U5[UR5:Xde[ U5US:XdeXl[R"5Ul [[U55GHn[S[[R"X5[R"US5- 55n/n[U5HnUR[!US:XaURUO UR"UR"SSUR$UR&UR(S95 XUS:wdMuUR[+SSUR,S 95 M URR[R."U65 GM" g) NrrrrSrX)rYrTrUrVrZr)r-r.r/r)rrrr^rminfeature_stridesrrl scale_headsromaxrBnplog2rprrrTrUrVr3r/ Sequential)rrkwargsrs head_length scale_headkrs rrFPNHead.__init__ss gt% 9- 917 9?#s4+;+;'<<<<?#q'9999.==?s?+,ABGGO./"''/!:L2MMNPKJ;'!!/0Av((+4==  !!%!% $ ./#%);;%% )*!+*.*<*<>?(    # #BMM:$> ?+-rc <URU5nURS"US5n[S[UR55H>nU[ URU"X$5UR SSSURS9-nM@ URU5nU$)NrrSrZr)r,r.r/) rrror^rr1rAr/r)rr~rrrss rrFPNHead.forward s  " "6 *!!!$QqT*q#d2234Af  #AD)\\!"%"00 22F5f% r)rr) r!r"r#r$rrrr%r&r's@rrrs@>rrc^\rSrSrSrS\"/SQSSS9\"/SQ/S Q/S QS S S \"SSS9S\"SSSS9S9 4U4SjjrSSjrSrSr U=r $) FPNSegmentoriz Packed Sementor Head Args: fpn_layer_indices: tuple of the indices of layers neck_cfg: dict of FPN params head_cfg: dict of FPNHead params )rX )rrr)rr]r`)rrrr)rrSrZrX)rr rBNT)rL requires_gradFrrr) rrrrrrrUr/rc2>[[U] 5 XlUSSn[R "[R "[R"XDSSS9[R"U5[R"5[R"XDSSS95[R"XDSSS9[R"5[R"SSS9/5Ul [S 0UD6Ul[S 0UD6UlUSUl[$R&R)5n[$R&R+S5 [$R&R-SSUR"S4S 9n[$R&R/U5 X`lg) NrrrZ)rr[r*rrXr,r)rrrfpn_layer_indicesrrlrConvTranspose2d BatchNorm2dGELUIdentity MaxPool2dpre_fpnrGfpn_neckrfpn_head NUM_CLASSESrrandom get_stateseedrandint set_statePALETTE)rrneck_cfghead_cfgwidthstatepaletters rrFPNSegmentor.__init__'s8& lD*,!2'*}} MM""5QqIu% ""5QqI     u1 E KKM LLQq 1 &   h +(+ $M2 ##% r))##As$2B2BA1F#G E" rc [R"U5nUR5nUn UcURc[R R 5n [R RS5 [R RSSURS4S9n[R RU 5 O URn[R"U5nURSUR:XdeURSS:Xde[UR5S:XdeSUs=:aS ::de eU RSUR:XdeU RSURS:wd U RSURS:wa*[XRSURS4SS S 5n U Sn [R "U SS 9n [R""U RSU RSS4[R$S 9n ['U5HupXX:HSS24'M U SSSS24n USU- -X--nUR)[R$5nUbSnU(a[R*"XU5 Ub[R,"X5 U(dU(d[.R1S5 ggg)aDraw `result` over `img`. Args: img (str or Tensor): The image to be displayed. result (Tensor): The semantic segmentation results to draw over `img`. palette (list[list[int]]] | np.ndarray | None): The palette of segmentation map. If None is given, random palette will be generated. Default: None win_name (str): The window name. wait_time (int): Value of waitKey param. Default: 0. show (bool): Whether to show the image. Default: False. out_file (str or None): The filename to write the image. Default: None. opacity(float): Opacity of painted segmentation map. Default 0.5. Must be in (0, 1] range. Returns: img (Tensor): Only if not `show` or `out_file` NrrrrXrrSrZrrTr)dtype.rFzMshow==False and out_file is not specified, only result image will be returned)mmcvimreadrdrrrrrrrrarrayrAr^r1rargmaxzerosuint8rzastypeimshowimwriterr)rimgresultrwin_nameshow wait_timeout_fileopacitysegr color_seglabelcolors r show_resultFPNSegmentor.show_resultVsh<kk#hhj ?||#  ++- r"))++As$:J:JA9N+O ##E*,,((7#}}Q4#3#3333}}Q1$$$7==!Q&&&7!c!!!!!yy|t///// 99Q<399Q< '399Q<399Q<+Gyy|SYYq\:D*C!fll3A&HHciilCIIaL!r+s(  99&ryy& IDryyD0E*EP\Z7\~7n7tR299Rr