T^(h2dZddlmZddlmZmZmZddlZddl m Z m Z ddl m Z mZmZddlmZmZmZmZdd lmZmZerddlZerddlZGd d e ZGd d edZGddeZdgZy)z Processor class for SAM. )deepcopy)ListOptionalUnionN) ImageInput VideoInput) ImagesKwargsProcessingKwargsProcessorMixin) AudioInput BatchEncodingPreTokenizedInput TextInput)is_tf_availableis_torch_availableceZdZUeeed<eeeeed<eeeeed<eeeeeed<eeed<y)SamImagesKwargssegmentation_maps input_points input_labels input_boxespoint_pad_valueN) __name__ __module__ __qualname__rr__annotations__rfloatintd/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/transformers/models/sam/processing_sam.pyrr%sW ++4U ,--4S ?++$tDK0122c]"r!rc$eZdZUeed<dddiiZy)SamProcessorKwargs images_kwargsrN)rrrrr _defaultsr r!r"r$r$-s"" s Ir!r$F)totalc eZdZdZdgZdZgdZfdZ ddddddee d ee e e e e e e fd eed eed ef d Z ddZdZ ddedej,d ej,fdZ ddZedZdZxZS) SamProcessora Constructs a SAM processor which wraps a SAM image processor and an 2D points & Bounding boxes processor into a single processor. [`SamProcessor`] offers all the functionalities of [`SamImageProcessor`]. See the docstring of [`~SamImageProcessor.__call__`] for more information. Args: image_processor (`SamImageProcessor`): An instance of [`SamImageProcessor`]. The image processor is a required input. image_processorSamImageProcessor)rrrrc`t|||jjd|_y)N longest_edge)super__init__r+size target_size)selfr+ __class__s r"r0zSamProcessor.__init__Ms) )//44^Dr!N)textaudiovideoimagesr5r6r7returnc |jtfdii||j|}|djdd}|djdd} |djdd} |djdd} |j|fi|d} | d} t | d r| j } |j|| | \}} } |j| | || | |d jd | } | S)z This method uses [`SamImageProcessor.__call__`] method to prepare image(s) for the model. It also prepares 2D points and bounding boxes for the model if they are provided. tokenizer_init_kwargsr%rNrrroriginal_sizesnumpy)rrr common_kwargsreturn_tensors)rrrr?r) _merge_kwargsr$'prepare_and_validate_optional_call_argspopr+hasattrr=_check_and_preprocess_points_normalize_and_convertget)r3r8r5r6r7argskwargs output_kwargsrrrrencoding_image_processorr<s r"__call__zSamProcessor.__call__QsV&+**  "$  ;d::DA  %_599.$O $_599.$O #O488M '8<<=NPTU#74#7#7 $ O,$  22BC >7 ++113N262S2S%%#3T3 / lK $(#>#> $ %%#(9==>NO+$?$  ('r!c t|tk7r/Dcgc]"}|j|j||d$c}n:t|D cgc]"\}} |j|j|| $c} }t fdDs||j ||\}t j|t j|}|t|t|k7r0|D cgc]$} |j|j| |dd&}} n;t||D cgc]$\} } |j|j| | d&}} } t j|}||dk(rAtj|}t|jdk7r|jdn|}nJ|dk(rEtj|}t|jdk7rtj|dn|}|jd |i|dk(rAtjtjd k7rjdnnJ|dk(rEtjtjd k7rtjdn|jd i||dk(rAtj|}t|jdk7r|jdn|}nJ|dk(rEtj|}t|jdk7rtj|dn|}|jd |i|Scc}wcc} }wcc} wcc} } w) Nrc3VK|] }|jdjk("yw)rN)shape).0pointrs r" z6SamProcessor._normalize_and_convert..s$Vu{{l1o&;&;;Vs&)T)is_bounding_boxptrtfrrr)len_normalize_coordinatesr2zipall_pad_points_and_labelsnparraytorch from_numpyrN unsqueezerUconvert_to_tensor expand_dimsupdate) r3rJr<rrrr?rrP original_sizeboxs ` r"rEz#SamProcessor._normalize_and_convertsd  #>"c,&77iu `eD//0@0@%XYIZ[  14L.0Q ,}//0@0@%W  VVV+151L1L$lO2.L,88L1L  #88L1L  ">"c+&66 +//0@0@#~VWGXjn/o /2+~.N*]//0@0@#}fj/k ((;/K  "%#..{; :=k>O>O:PTU:Uk33A6[f 4' 22;? @CKDUDU@VZ[@[bnn[!dp $ + +^\,J K  #%$// = dp $ + +^\,J K''A  $ s'M&'M+9)M14)M6c t|Dcgc]}|jdc}}g}t|D]\}}|jd|k7r^tj|tj ||jdz df|zgd}tj |||g||<|j ||}||fScc}w)zh The method pads the 2D points and labels to the maximum number of points in the batch. r)axis)maxrN enumerater\ concatenatezerosappend)r3rrrrPexpected_nb_pointsprocessed_input_pointsis r"r[z#SamProcessor._pad_points_and_labelss!l!KU%++a.!KL!#!,/ 1HAu{{1~!33BHH&85;;q>&I1%MNQ``ahi#%))LOo=N"O Q " ) )% 0  1. \))"LsCr2coordsc|\}}|jj||\}}t|jt}|r|j ddd}|d||z z|d<|d||z z|d<|r|j dd}|S)z~ Expects a numpy array of length 2 in the final dimension. Requires the original image size in (H, W) format. )r.rg).r).rTrV)r+_get_preprocess_shaperastyperreshape) r3r2rqrdrRold_hold_wnew_hnew_ws r"rXz#SamProcessor._normalize_coordinatess % u++AA-^iAj u&!((/ ^^B1-F55=9v55=9v ^^B*F r!c|{t|dr|jj}t|trt|dts t d|Dcgc]}t j|}}nd}|{t|dr|jj}t|trt|dts t d|Dcgc]}t j|}}nd}|t|dr|jj}t|tr)t|dtrt|ddts t d|Dcgc]4}t j|jt j6}}nd}|||fScc}wcc}wcc}w)a8 Check and preprocesses the 2D points, labels and bounding boxes. It checks if the input is valid and if they are, it converts the coordinates of the points and bounding boxes. If a user passes directly a `torch.Tensor`, it is converted to a `numpy.ndarray` and then to a `list`. Nr=rz7Input points must be a list of list of floating points.z-Input labels must be a list of list integers.z>Input boxes must be a list of list of list of floating points.) rCr=tolist isinstancelist ValueErrorr\r]rufloat32)r3rrr input_pointlabelres r"rDz)SamProcessor._check_and_preprocess_pointss  #|W-+113::< lD1LQROUY9Z !Z[[EQRkBHH[1RLRL  #|W-+113::< lD1LQROUY9Z !PQQ9EFBHHUOFLFL  "{G,)//188: {D1!+a.$7!+a."3T: !abbGRS288C=// ;SKSK\;669SGTsF6F;39Gcj|jj}ttj |SN)r+model_input_namesr~dictfromkeys)r3image_processor_input_namess r"rzSamProcessor.model_input_names.s)&*&:&:&L&L#DMM"=>??r!c:|jj|i|Sr)r+post_process_masks)r3rGrHs r"rzSamProcessor.post_process_masks3s 6t##66GGGr!r)NNNrSr&)F)NNN)rrr__doc__ attributesimage_processor_classoptional_call_argsr0rrrrrrr r rrKrEr[rr\ndarrayrXrDpropertyrr __classcell__)r4s@r"r*r*6s $$J/E (,9(im&*&*9($9(uY(94 ?DQbLccde9( #9( #9( 9(~L(\*"TY(*  . -7^@@Hr!r*)rcopyrtypingrrrr=r\ image_utilsrr processing_utilsr r r tokenization_utils_baser rrrutilsrrr^ tensorflowrUrr$r*__all__r r!r"rst((1NN^^8#l#)~H>~HB  r!