^j28dZddlmZddlZddlmZddlmZddlm Z m Z ddl m Z m Z mZmZmZmZmZmZdd lmZmZdd lmZmZmZerd d lmZerdd lmZmZddl m!Z"dZ#de fdZ$ ddZ% ddZ&GddedZ'eGddeZ(dgZ)y)z$Image processor class for SuperGlue.) TYPE_CHECKINGN)TorchvisionBackend) BatchFeature)group_images_by_shapereorder_images) ImageInput ImageTypePILImageResamplingSizeDictget_image_type is_pil_imageis_valid_imageto_numpy_array) ImagesKwargsUnpack) TensorTypeauto_docstringis_vision_available)SuperGlueKeypointMatchingOutput)Image ImageDraw) functionalct|xsCt|xr6t|tjk7xrt |j dk(S)Nrrrr r PILlenshapeimages /var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/superglue/image_processing_superglue.py_is_valid_imager#2sF   ub."79=="HbSQVQ\Q\M]abMbimagescd}dt|trQt|dk(rtfd|Dr|Stfd|Dr|Dcgc] }|D]}| c}}St |cc}}w)N)z-Input images must be a one of the following :z - A pair of PIL images.z - A pair of 3D arrays.z! - A list of pairs of PIL images.z - A list of pairs of 3D arrays.ct|xsCt|xr6t|tjk7xrt |j dk(S)z$images is a PIL Image or a 3D array.rrr s r"r#z8validate_and_format_image_pairs.._is_valid_imageAsGE" 5 ! fnU&;y}}&L fQTUZU`U`QaefQf r$c3.K|] }|ywN.0r!r#s r" z2validate_and_format_image_pairs..Hs#Q_U%;#Qc3K|]:}t|txr$t|dk(xrtfd|D<yw)r(c3.K|] }|ywr*r+r,s r"r.z..MsCuOE*Cr/N) isinstancelistrall)r- image_pairr#s r"r.z2validate_and_format_image_pairs..JsN  z4 ( DJ1$ DC CC D sAA)r2r3rr4 ValueError)r% error_messager5r!r#s @r"validate_and_format_image_pairsr88sM &$ v;! #Q&#Q QM  %   -3Kj KuEKEK K ] ##LsA3c $|jdks#|j|jdk(rdnddk(rytj|ddddddf|ddddddfk(xr.tj|ddddddf|ddddddfk(S)zAChecks if an image is grayscale (all RGB channels are identical).rrrT.Nr()ndimrtorchr4r s r" is_grayscaler<Ts zzA~%**/QqAQF 99U31a<(E#q!Q,,?? @ UYY c1aluS!Q\22Fr$cJt|r|Stj|dS)a Converts an image to grayscale format using the NTSC formula. Only support torch.Tensor. This function is supposed to return a 1-channel image, but it returns a 3-channel image with the same value in each channel, because of an issue that is discussed in : https://github.com/huggingface/transformers/pull/25786#issuecomment-1730176446 Args: image (torch.Tensor): The image to convert. r)num_output_channels)r<tvFrgb_to_grayscaler s r"convert_to_grayscalerA_s$E   1 ==r$ceZdZUdZeed<y)SuperGlueImageProcessorKwargsz do_grayscale (`bool`, *optional*, defaults to `self.do_grayscale`): Whether to convert the image to grayscale. Can be overridden by `do_grayscale` in the `preprocess` method. do_grayscaleN)__name__ __module__ __qualname____doc__bool__annotations__r+r$r"rCrCrs r$rCF)totalcxeZdZeZej ZdddZdZ dZ dZ dZ dZ dZdeeffd Zed edeed effd Zd ed efd Z d"d eddededddedededzdeezdzded efdZ d#dddeeezded eeeej@ffdZ!deeeej@fd edfd Z"d!Z#xZ$S)$SuperGlueImageProcessorii)heightwidthFTgp?Nkwargsc $t|di|y)Nr+)super__init__)selfrP __class__s r"rSz SuperGlueImageProcessor.__init__s "6"r$r%returnc $t||fi|Sr*)rR preprocess)rTr%rPrUs r"rXz"SuperGlueImageProcessor.preprocesssw!&3F33r$c :|j|}t|Sr*) fetch_imagesr8)rTr%rPs r"_prepare_images_structurez1SuperGlueImageProcessor._prepare_images_structures ""6*.v66r$ torch.Tensor do_resizesizeresamplez7PILImageResampling | tvF.InterpolationMode | int | None do_rescalerescale_factordisable_groupingreturn_tensorsrDc  (t||\} } i} | jD] \}}|r|j|||}|| |<"t| | }t||\} } i} | jD]+\}}|r|j ||}| r t |}|| |<-t| | }t dt|dDcgc] }|||dz }}|Dcgc]}tj|d}}td|i|Scc}wcc}w)N)rb)r^r_rr()dim pixel_values)data tensor_type) ritemsresizerrescalerArangerr;stackr)rTr%r]r^r_r`rarbrcrDrPgrouped_imagesgrouped_images_indexprocessed_images_groupedrstacked_imagesresized_imagesprocessed_imagesi image_pairspair stacked_pairss r" _preprocessz#SuperGlueImageProcessor._preprocesssS0EV^n/o,,#% %3%9%9%; = !E>!%^$QY!Z.< $U + =((@BVW/D^fv/w,,#% %3%9%9%; = !E>!%nn!M!5n!E.< $U +  =**BDXY=B!SIYEZ\]<^_q'AE2_ _?JJdTq1J J.-!@n]]`Ks D Doutputsr target_sizes thresholdc|jjdt|k7r tdt d|Ds tdt |t r,tj||jj}n1|jddk7s|jddk7r td|}|jj}||jdjddddz}|jtj}g}t!|j||j"d d df|j$d d dfD]v\}}} } |ddkD} |ddkD} |d| } |d| }| | }| | }||kD|dkDz||jdkz}| |}|||}||}|j'|||d x|S) a Converts the raw output of [`SuperGlueKeypointMatchingOutput`] into lists of keypoints, scores and descriptors with coordinates absolute to the original image sizes. Args: outputs ([`SuperGlueKeypointMatchingOutput`]): Raw outputs of the model. target_sizes (`torch.Tensor` or `list[tuple[tuple[int, int]]]`, *optional*): Tensor of shape `(batch_size, 2, 2)` or list of tuples of tuples (`tuple[int, int]`) containing the target size `(height, width)` of each image in the batch. This must be the original image size (before any processing). threshold (`float`, *optional*, defaults to `0.0`): Threshold to filter out the matches with low scores. Returns: `list[Dict]`: A list of dictionaries, each dictionary containing the keypoints in the first and second image of the pair, the matching scores and the matching indices. rzRMake sure that you pass in as many target sizes as the batch dimension of the maskc38K|]}t|dk(yw)r(N)r)r- target_sizes r"r.zISuperGlueImageProcessor.post_process_keypoint_matching..sI[3{#q(IszTEach element of target_sizes must contain the size (h, w) of each image of the batch)devicerr(N) keypoints0 keypoints1matching_scores)maskrrr6r4r2r3r;tensorr keypointscloneflipreshapetoint32zipmatchesrappend)rTryrzr{image_pair_sizesrresults mask_pairkeypoints_pairrscoresmask0mask1rrmatches0scores0 valid_matchesmatched_keypoints0matched_keypoints1rs r"post_process_keypoint_matchingz6SuperGlueImageProcessor.post_process_keypoint_matchings, <<  a C $5 5qr rILIIst t lD )$||LATATU !!!$)\-?-?-Ba-G j , %%++-  0 5 5b 9 A A"aA NN LL- := LL)W__QT%:G>.1D)4\)B)I)I!)L &L,)4\)B)I)I!)L &L,VYlL, TeHfW R [+{N7  +{V/C[Q  kAo{Q aQ\_`Q`aho p  6)A-{Q f@TWX@XZehiZij   NN> *7 +8A=Ks HHcNtdd|z z}td|z}d}|||fS)zMaps a score to a color.rr)int)rTscorergbs r"rz"SuperGlueImageProcessor._get_color=s3 q5y! " e  !Qwr$)T)g)%rErFrGrC valid_kwargsr BILINEARr_r^default_to_squarer]r`ra do_normalizerDrrSrr rrXr[r3rIr floatstrrrxtupledictr;Tensorrrr __classcell__)rUs@r"rMrM{s0L!**HC (DIJNLL#(E!F#44v>[7\4am4477  7&")^^$)^)^ )^ L )^  )^)^+)^j(4/)^)^ )^^ B2B!4;.B B d3 $% & BH5#'tC,='>"?5 m  5nr$rM)r!r\)r!r\rVr\)*rHtypingrr;image_processing_backendsrimage_processing_utilsrimage_transformsrr image_utilsr r r r r rrrprocessing_utilsrrutilsrrrmodeling_supergluerrrrtorchvision.transforms.v2rr?r#r8r<rArCrM__all__r+r$r"rs+ ;2E   5C$7 $J$8 > >>&LF0FFR % %r$