Zj7FSSKrSSKrSSKJrJr SSKJrJrJ r J r SSK J r J r SSKJrJrJr SSKJr SSKJrJrJr \"5(aSS KJr \R6"\5rS rS rS r S r!Sr"S\#SS\#\$4Sjr%S\$S\&\$\$44Sjr'"SS\ SS9r("SS\5r)S/r*g)N) BatchFeature get_size_dict)IMAGENET_STANDARD_MEANIMAGENET_STANDARD_STDPILImageResamplingSizeDict)Unpack VideosKwargs) TensorTypeis_torchvision_availablelogging)BaseVideoProcessor) VideoMetadatagroup_videos_by_shapereorder_videos) functionalzYou are a helpful language and vision assistant. You are able to understand the visual content that the user provides, and assist the user with a variety of tasks using natural language.zgYou are provided the following series of {frame_count} frames from a {video_duration} [H:MM:SS] video. z z Frame from {timestamp}:ivideos torch.Tensorreturnc[S5=pUH.nUR5SSupE[XA5n[XR5nM0 X4$)z@ Get the maximum height and width across all videos in a batch. z-infN)floatsizemax)r max_height max_widthvideoheightwidths څ/root/GenerationalWealth/GenerationalWealth/venv/lib/python3.13/site-packages/transformers/models/smolvlm/video_processing_smolvlm.pyget_max_height_widthr"/sM#6]*J RS) , )   ""resolution_max_sidec4UR5SSup#[[U5nUc [X#5OUnX2- nX2:aUn[ X4- 5nUS-S:waUS- nO"X#:aUn[ X$-5nUS-S:waUS- n[US5n[US5nX#4$)a Get the output size of the video after resizing given a dictionary specifying the max and min sizes. Args: video (`np.ndarray`): Video to resize. resolution_max_side (`int`): The longest edge of the video will be resized to this value. The shortest edge will be resized to keep the input aspect ratio. Returns: The output size of the video after resizing. rNr)rminMAX_IMAGE_SIZErint)rr$rr aspect_ratios r!get_resize_output_image_sizer,;sJJL%MFn.AB0C0K#f,Qd>L #U)* A:? aKF $F)* 19> QJE ^F qME =r#c*\rSrSr%\\\4\S'Srg)SmolVLMVideoProcessorInitKwargscmax_image_sizeN) __name__ __module__ __qualname____firstlineno__dictstrr*__annotations____static_attributes__r1r#r!r.r.cscN"r#r.F)totalc^\rSrSr\R rSS0rSS0r\ r \ r Sr SrSrSrSrSr\rSS/rS \\4U4S jjrS*S S S\SSS\SS 4 U4SjjjrS+S S S\\\4S\S\S\4 SjjrS,S\S\S -S\\-S -S\S -4Sjjr S-S\!S S\S \S\SSS!\S"\S#\S$\S%\\!\-S -S&\\!\-S -S'\"\#-S -4S(jjr$S)r%U=r&$).SmolVLMVideoProcessorg longest_edgeiilTF pixel_valuespixel_attention_maskkwargsc >[TU]"S0UD6 SU;aSU;a USUSS'SU;a:USSUlUSSUl[ USSUR S9Ulgg)Nrvideo_sampling video_size max_framesfps)default_to_squarer1)super__init__ num_framesrFrrGr)selfrA __class__s r!rISmolVLMVideoProcessor.__init__vs "6" V  0F :5;F^F# $\ 2 v %$%56|DDO./6DH%f-=&>|&L`d`v`vwDI &r#Nrrrresamplez7PILImageResampling | tvF.InterpolationMode | int | None antialiasrc >UR(a[UURS9nOJUR(a*UR(aURUR4nO[ SUS35e[ TU]U[USUSS9X4S9n[URSURSS9n[ TU]XX4S9nU$) a Resize an video to `(size["height"], size["width"])`. Args: video (`torch.Tensor`): Video to resize. size (`SizeDict`): Dictionary in the format `{"height": int, "width": int}` specifying the size of the output video. resample (`PILImageResampling` or `InterpolationMode`, *optional*, defaults to `InterpolationMode.BILINEAR`): Resampling filter to use when resizing the video. Returns: `torch.Tensor`: The resized video. )r$zHSize must contain 'height' and 'width' keys, or 'longest_edge' key. Got .rr')rr )rNrOr>) r>r,rr ValueErrorrHresizer r0) rKrrrNrOrAnew_sizemax_sizerLs r!rSSmolVLMVideoProcessor.resizes(   4$($5$5H[[TZZ TZZ0Hghlgmmnop p 88A;hqkBX  4#6#6~#FdNaNabpNqruW r# padded_sizemax_num_framesfillreturn_pixel_maskcUR5SSnUSUS- nUSUS- nX1RS- n US:dUS:a[SUSUS35eXb:waSUSUSSSU /n [R"XUS9nSn U(a@[ R "US SSS2SS24[ RS 9n SU S SUS2SUS24'X4$) aPads the sample with empty video to the padded_size Args: video (`torch.Tensor`): Video to pad. padded_size (`tuple[int, int]`): Height and width to pad. max_num_frames (`int`): The maximum number of frames to which video will be padded. fill (`int`, *optional*): The value to use for the padding. return_pixel_mask (`bool`, *optional*, defaults to `True`): Whether to return a pixel mask. rNrr'zzPadding dimensions are negative. Please make sure that the padded size is larger than the original size. Got padded size: z, original size: rQ)rY.dtype)rshaperRtvFpadtorch zeros_likeint64) rKrrWrXrYrZ original_sizepadding_height padding_width padding_framepadding pixel_masks r!r`SmolVLMVideoProcessor.pads* RS) $Q-*::#Aq)99 &Q7 1  233>-?PQ^P__`b   '-NAq!]SGGGE6E ))%Q1 *=U[[QJFGJs.mA..0B-2B0BB C  r#metadatarJrF skip_secsc Ub[USS5c [S5eUbUO URnUbUO URnURn[ [ X1S-55n[Xr5nUS:aSnSn US- n US:a4USSU-- X#-:a#[ XAS-5n [ XdUS-- 5n [SU 5n [XS- 5n X:aSUS- p[R"XU[ S9n [R"U 5n U $) a Video sampling function which: - Uses `num_frames` (if provided) or calculates it from `fps` and metadata. - Applies a basic center-skip if fewer frames than available, otherwise optionally skips `skip_secs` from both the start and end. - Uniformly samples the desired number of frames between the start and end indices. Args: metadata (`VideoMetadata`): Metadata of the video containing information about total duration, fps and total number of frames. num_frames (`int`, *optional*): Maximum number of frames to sample. Defaults to `self.num_frames`. fps (`int` or `float`, *optional*): Target frames to sample per second. Defaults to `self.fps`. skip_secs (`float`, *optional*, defaults to `1`): Number of seconds to skip from the start and end if the video is long enough. Returns: np.ndarray: Indices to sample video frames. NrFzAsked to sample frames per second but no video metadata was provided which is required when sampling in SmolVLM. Please pass in `VideoMetadata` object or set `do_sample_frames=False`durationr'rr&r\) getattrrRrJrFtotal_num_framesr*roundr(rnplinspaceunique) rKrkrJrFrlrArpestimated_framesdesired_frames start_idxend_idxindicess r! sample_frames#SmolVLMVideoProcessor.sample_framess7:  wx=EX  $.#9Zt _c$((#44uSJ+?%?@A-: A N "Q& q=hz2Q]BzGWXI78I*%-HHIG9% g!34  !"$4q$8w++i.L))G$r#rdo_convert_rgb do_resize do_rescalerescale_factor do_normalizedo_pad image_mean image_stdreturn_tensorsc [U5up0nUR5H:unnU(aURU5nU(aURUXES9nUUU'M< [ UU5n[U5up0nUR5HunnUR UXgXU 5nUUU'M! [ UU5nU (a~[ U5n[SU55n[U5up0n0nUR5H$unnURUUUS9unnUUU'UUU'M& [ UU5n[ UU5nSU0nU (a$U (aU b[R"WSS9OWUS'[UU S9$) N)rrNc38# UHn[U5v M g7fN)len).0rs r! 4SmolVLMVideoProcessor._preprocess..>s J9IU9Is)rWrXr?r)dimr@) tensor_type) ritemsconvert_to_rgbrSrrescale_and_normalizer"rr`rastackr)rKrr|r}rrNr~rrrrrrrAgrouped_videosgrouped_videos_indexresized_videos_groupedr^stacked_videosresized_videosprocessed_videos_groupedprocessed_videospad_sizerXprocessed_padded_mask_grouped padded_masksr@datas r! _preprocess!SmolVLMVideoProcessor._preprocesss 0EV/L,!#%3%9%9%; !E>!%!4!4^!D!%^$!Z,: "5 ) &< ((>@TU/D^/T,#% %3%9%9%; !E>!77 LV_N/= $U + &< **BDXY +,<=H J9I JJN3HIY3Z 0N,. )') $)7)=)=)?%~/3xx"080, 3A(/7C-e4 *@ ..FH\] #12OQe#f  01 n8 0a8) ' ( Dn==r#)rFrJr)NT)rT)NNr'r)'r2r3r4r5rLANCZOSrNrr0rrrrr}r~rr|rdo_sample_framesr. valid_kwargsmodel_input_namesr rIr boolrStupler*r`rrrzlistr7r rr9 __classcell__)rLs@r!r<r<gs!))H G $D$c*N'J%IIJLN F2L')?@ x(G!H x OS (((L (  ( ((^"& )!)!38_)! )!  )!  )!\"&"& ! ??$J?5[4  ? : ?\37=>^$=>=> =>  => L =>=>=>=>=>DK'$.=>4;&-=>j(4/=>=>r#r<)+numpyrrraimage_processing_utilsrr image_utilsrrrr processing_utilsr r utilsr r rvideo_processing_utilsr video_utilsrrrtorchvision.transforms.v2rr_ get_loggerr2loggerDEFAULT_SYSTEM_MESSAGEDEFAULT_VIDEO_INTRODEFAULT_MEDIA_OUTTROFRAME_TIMESTAMP_MESSAGEr)rr*r"rr,r.r<__all__r1r#r!rs  A 5BB8OO;   H %Vn5 #n!5 #$s) #%% 38_%P#l%#n>.n>b # #r#