9i&4SrSSKrSSKrSSKrSSKJr SSKJr SSK J r SSK J r SSKJ r SSKJrJr SSKJr S S KJr S rS rS rS1S\R0S\4SjjrS2S\R0S\S\S\S\4 SjjrS3S\R0S\R0S\4SjjrS4S\R0S\R0S\R0S\R0S\R0S\R0S\4SjjrS5S\R0S\R0S\R@S \R@S!\R0S"\S#\!S$\"S%\"S&\4S'jjr#S(\$S)\"S*\"S+\"4S,jr%S(\$S-\R@S.\&S/\"S%\"S&\S)\"S*\"S+\"4S0jr'g)6zx Part of the implementation is borrowed and modified from LaMa, publicly available at https://github.com/saic-mdal/lama N)gaussian_blur2d)resizeerosion) functional)SGDAdam)tqdm)FFCResnetBlockc [U[R5(aURU5$[R "U5(aURU5$[U[ [45(aUVs/sHn[X!5PM sn$[U[5(a/UR5VVs0sHup4U[XA5_M snn$[S[U535es snfs snnf)NzUnexpected type ) isinstancennModuletotorch is_tensortuplelistmove_to_devicedictitems ValueErrortype)objdeviceelnamevals p/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/modelscope/models/cv/image_inpainting/refinement.pyrrs#ryy!!vvf~ svvf~#t}%%589Srr*S99#tCF99;O;idnS11;OO 'S {3 44:Os <C/<C4c*X-S:XaU$X-S-U-$)Nrr )xmods r ceil_modulor%s!w!| HqLC cURup#pE[XA5n[XQ5n[R"USXu- SXd- 4SS9$)Nrreflect)padmode)shaper%Fr))imgr$ batch_sizechannelsheightwidth out_height out_widths r pad_tensor_to_modulor4%sM*-))'J&V)JE'I 55  !1j&9 :  r&imdownsizecUc$URSS-URSS-4nURSS:XdS5e[USSS9n[R"XSS S 9nU$) zdownscale the imager z5Expected shape for the input to be (n,3,height,width)r;?r= kernel_sizesigmabilinearFsizer* align_cornersr+rr, interpolate)r5r6s r _pyrdownrG/s|HHQK1$bhhqkQ&67 88   IHI  z BB rz OB Ir&maskeps blur_maskround_upcXUc$URSS-URSS-4nURSS:XdS5eUSLa"[USSS9n[R"XS S S 9nO[R"XS S S 9nU(aSXU:'S XU:'U$SXS U- :'S XS U- :'U$)a2downscale the mask tensor Parameters ---------- mask : torch.Tensor mask of size (B, 1, H, W) downsize : tuple, optional size to downscale to. If None, image is downscaled to half, by default None eps : float, optional threshold value for binarizing the mask, by default 1e-8 blur_mask : bool, optional if True, apply gaussian filter before downscaling, by default True round_up : bool, optional if True, values above eps are marked 1, else, values below 1-eps are marked 0, by default True Returns ------- torch.Tensor downscaled mask r8r9r z5Expected shape for the input to be (n,1,height,width)Tr:r<r>rAFrBrr=rE)rHr6rIrJrKs r _pyrdown_maskrM:s4JJqMQ& 1 (:; ::   IHI DtzJ}} jG}} jGS[CZ K#$S3Y !"C#I  Kr&ekernelcJUb[X5nSXSU- :'SXSU- :'U$)z(erode the mask, and set gray pixels to 0r r=rr)rHrNrIs r _erode_maskrPhs;t%"#S3Y !"C#I  Kr&predpred_downscaledrefmask_downscaledimageon_predc [R"[R"XS:XSS:- 55nU(a:U[R"[R"XS:X$S:- 55- nU$)zAl1 loss on src pixels, and downscaled predictions if on_pred=True:0yE>)rmeanabs)rQrRrSrHrTrUrVlosss r _l1_lossr\sss ::eiiD[ 1E+4F FG HD  IIo&=>t345 67 7 Kr& forward_front forward_rears ref_lower_res orig_shapedevices scale_indn_iterslrc \USU- -n [R"X/SS9n URSSSS5nUbUR5n[R"5 U"U 5upSSS5 UR US5n[R "[R"[RS5R[55R5n U R US5n UR US5nW R5R US5W R5R US5pSuU l U l [X/U S 9n[[!U5S S 9nUGHsnUR#5 X4n[%U5H[unnU"U5nU['U5S- :a9UunnUR UUS-5UR UUS-5nnUU4nMYUnM] Uc O0n[)WSS2SS2SUS2SUS245n[+USS2SS2SUS2SUS24S S S 9n[-UU S 9nURSSSS5n[/UUUUUUSS9US'[1UR355nUR5SR7US-UUR9555 UUS- :dGMPUR;5 UR=5 AAAGMv UW-SU- U--nUR5R?5nU$!,(df  GN=f)aPerforms inference with refinement at a given scale. Parameters ---------- image : torch.Tensor input image to be inpainted, of size (1,3,H,W) mask : torch.Tensor input inpainting mask, of size (1,1,H,W) forward_front : nn.Module the front part of the inpainting network forward_rears : nn.Module the rear part of the inpainting network ref_lower_res : torch.Tensor the inpainting at previous scale, used as reference image orig_shape : tuple shape of the original input image before padding devices : list list of available devices scale_ind : int the scale index n_iters : int, optional number of iterations of refinement, by default 15 lr : float, optional learning rate, by default 0.002 Returns ------- torch.Tensor inpainted image r )dimr9N)rhr)TT)rdF)leave)rJrK)rNT)rVms_l1z8Refining scale {} using scale {} ...current loss: {:.4f}) rcatrepeatdetachno_gradr from_numpycv2getStructuringElement MORPH_ELLIPSEastypeboolfloat requires_gradr r range zero_grad enumeratelenrGrMrPr\sumvaluesset_descriptionformatitembackwardstepcpu)rUrHr]r^r_r`rarbrcrd masked_imagez1z2rN optimizerpbaridi input_featidd forward_rear output_featmidz1midz2rQlossesrRrTr[ inpainteds r _inferrs*PAH%L99l1q9L ;;q!Q "D %,,. |,  7772; D !!#"3"3"* ,,2F4L::?%' jj%G HHWR[ !E YY[^^GAJ ' )C)3&Bb&bX"%I ge ,DX !*=!9 C&z2KS\A%%* u$xxa(89588C!G$<&u#U^ "":  "41nz!}nnz!}n(L#MN' BQBAA6 7&owG)00Aq!<"      w6==?#  F M MA y$))+ 7 8 1  MMO NN UXt q4x500I  "&&(I { s  L L+batchmin_side max_scales px_budgetcrUSRSS:XdS5eUSupEUSR5USR5pTUSSSU2SU24nUSSSU2SU24nXE-U:at[R"U[ XE-5- 5nXEp[ XH-5[ XX-5pT[ S X4S XE4S 35 [XdU4S S S9n[XtU4S S S9nSXwS:'[XE5n [S[ [[S[R"X- 5555-U5n /n /nU RU5 URU5 [U S- 5HAn[U S5n[US5nU RU5 URU5 MC U SSS2USSS24$)aBuild the image mask pyramid Parameters ---------- batch : dict batch containing image, mask, etc min_side : int minimum side length to limit the number of scales of the pyramid max_scales : int maximum number of scales allowed px_budget : int the product H*W cannot exceed this budget, because of resource constraints Returns ------- tuple image-mask pyramid in the form of list of images and list of masks rUrr z(refiner works on only batches of size 1! unpad_to_size.NrHz2Original image too large for refinement! Resizing z to z...rAF) interpolationrDrXrg)r+rnpsqrtruintprintrminroundmaxlog2appendrwrGrM)rrrrhwrUrHratioh_origw_origbreadthn_scales ls_imagesls_masks_image_pmask_ps r _get_image_mask_pyramidrs* >     <;<   !DA Q499;! q 'N3BQB; 'E =bqb"1" %Duy E!%L0119~s19~1 @&@QQUWXV[U\\_ `  q65J a& %ID[!iG1s5Q0B(C!DEFFHIH U OOD 8a< 9R=)x|,! ! TrT?HTrTN **r& inpaintergpu_idsmoduloc FURnUR(aeUR(aeUR(deUR SS5R S5V s/sHn U R 5(dMSU 3PM! nn Sn Sn Sn [[URR55HFn [URRU [5(a U S- n Sn M8U (aMAU S- n MH U [U5-nUVs/sHn[R"U5PM nnURRSU nURUS5 /n[[U55HnU[U5S- :a8URURRU UU--U UUS---5 O.URURRU UU--S 5 UURUU5 M [!XUU5unnS n[#[%UU55HununnUR&S S n[)UU5n[)UU5nS UUS :'S UUS :'[+UUS5[+UUS5nnUb[+UUS5n[-UUUUUUUUXE5 nUS S 2S S 2S US2S US24nUR/5R15nUR/5R15nM U$s sn fs snf)aRefines the inpainting of the network Parameters ---------- batch : dict image-mask batch, currently we assume the batchsize to be 1 inpainter : nn.Module the inpainting neural network gpu_ids : str the GPU ids of the machine to use. If only single GPU, use: "0," modulo : int pad the image to ensure dimension % modulo == 0 n_iters : int number of iterations of refinement for each scale lr : float learning rate min_side : int all sides of image on all scales should be >= min_side / sqrt(2) max_scales : int max number of downscaling scales for the image-mask pyramid px_budget : int pixels budget. Any image will be resized to satisfy height*width <= px_budget Returns ------- torch.Tensor inpainted image of size (1,3,H,W)  ,zcuda:rFr TNr8r=rXgrg)modeltrainingadd_noise_kwargs concat_maskreplacesplitisdigitrwrz generatorrr rrrrrryzipr+r4rrrmr)rrrrrcrdrrrgpuidn_resnet_blocksfirst_resblock_indfound_first_resblockidlresblocks_per_gpugpu_idrar]r^rrrimage_inpaintedidsrUrHr`s r refine_predictr-sQ>I!!! !))) )   &-__S"%=%C%CC%H%HE ==? %w%H O S,,2234 i))//4n E E q O#' %% ! #  5 (3w<729:'u||F#'G:''--a0BCMWQZ MS\" W! !  ##))*<,=-0-2+22D,=q,I3JK L  ##))*<,=,E+F+GH I cgcl+#2%:2;=IxO'Ix(@A]eT[[_ $UF3#D&1 TT\TD[$UGAJ7 '!*:t  &,_gbkJO m]!0*gs!(.*!QAA*NO ""${{}  "!B$ q;s)L L L)N)NrXTT)NrX)T)rhgMb`?)(__doc__rpnumpyrrtorch.nnrkornia.filtersrkornia.geometry.transformrkornia.morphologyrrr, torch.optimrr r modules.ffcr rr%r4TensorrrGrurtrMrPr\rrrrrrstrrr"r&r rs(  *,%$!' 5 %)#$(#' + +!++"+! +^)-!ell"" 5<< #ll ,, << $ll  LL   0k%,,kk))k))k ,, k  k  kkkk\9+49+39+C9+'*9+x\$\299\s\\),\27\CF\"\/2\r&