o hd @sddlZddlZddlmZddlmZmZmZmZm Z m Z m Z ddl Z ddlZddlmZddlmZmZmZmZmZmZmZmZddlmZmZmZmZmZm Z ertddl!Z"ddl#Z"e$e$e"j%j&e$dkrqe"j'j(Z)ne"j'Z)er}er}ddl*Z*e d e j+d ed ee j+ed fZ,Gd d d eZ-d dZ.ddZ/ddZ0ddZ1de j+de2fddZ3d8de4dee,fddZ5de j+fddZ6 d9de j+de e e4e e4d ffde-fd!d"Z7 d9de j+d#e e e-e8fde4fd$d%Z9d9de j+d&e-de e4e4ffd'd(Z:d)ee8e ee ffde2fd*d+Z;d)ee8e ee ffde2fd,d-Zd9de e8d fd3e e?dd fd4d5Z@Gd6d7d7ZAdS):N)BytesIO) TYPE_CHECKINGDictIterableListOptionalTupleUnion)version) ExplicitEnum is_jax_tensor is_tf_tensoris_torch_availableis_torch_tensoris_vision_availablerequires_backendsto_numpy)IMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STDIMAGENET_STANDARD_MEANIMAGENET_STANDARD_STDOPENAI_CLIP_MEANOPENAI_CLIP_STDz9.1.0zPIL.Image.Imagez torch.Tensorc@seZdZdZdZdS)ChannelDimensionchannels_first channels_lastN)__name__ __module__ __qualname__FIRSTLASTr"r"N/var/www/html/ai/venv/lib/python3.10/site-packages/transformers/image_utils.pyr@srcCsto t|tjjSN)r isinstancePILImageimgr"r"r# is_pil_imageEsr*cCs8tr t|tjjpt|tjpt|pt|pt|Sr$) rr%r&r'npndarrayrrr r(r"r"r#is_valid_imageIs r-cCs:t|ttfr|D] }t|sdSq dSt|sdSdS)NFT)r%listtuple valid_imagesr-)imgsr)r"r"r#r0Ssr0cCst|ttfr t|dSdS)NrF)r%r.r/r-r(r"r"r# is_batched_s r2imagereturncCs,|jtjkrdSt|dkot|dkS)zV Checks to see whether the pixel values have already been rescaled to [0, 1]. Frr )dtyper+uint8minmax)r3r"r"r#is_scaled_imagees r9expected_ndimscCst|r|St|tjjr|gSt|r<|j|dkr!t|}|S|j|kr+|g}|Std|dd|d|jdtdt|d)a Ensure that the input is a list of images. If the input is a single image, it is converted to a list of length 1. If the input is a batch of images, it is converted to a list of images. Args: images (`ImageInput`): Image of images to turn into a list of images. expected_ndims (`int`, *optional*, defaults to 3): Expected number of dimensions for a single input image. If the input image has a different number of dimensions, an error is raised. r z%Invalid image shape. Expected either z or z dimensions, but got z dimensions.ztInvalid image type. Expected either PIL.Image.Image, numpy.ndarray, torch.Tensor, tf.Tensor or jax.ndarray, but got .) r2r%r&r'r-ndimr. ValueErrortype)imagesr;r"r"r#make_list_of_imagesps*  rAcCs@t|s tdt|trt|tjjrt|St |S)NzInvalid image type: ) r-r>r?rr%r&r'r+arrayrr(r"r"r#to_numpy_arrays  rC num_channels.cCs|dur|nd}t|tr|fn|}|jdkrd\}}n|jdkr&d\}}ntd|j|j||vr8tjS|j||vrBtjStd)a[ Infers the channel dimension format of `image`. Args: image (`np.ndarray`): The image to infer the channel dimension of. num_channels (`int` or `Tuple[int, ...]`, *optional*, defaults to `(1, 3)`): The number of channels of the image. Returns: The channel dimension of the image. Nr r:r:)rz(Unsupported number of image dimensions: z(Unable to infer channel dimension format)r%intr=r>shaperr r!)r3rD first_dimlast_dimr"r"r#infer_channel_dimension_formats    rLinput_data_formatcCsF|durt|}|tjkr|jdS|tjkr|jdStd|)a Returns the channel dimension axis of the image. Args: image (`np.ndarray`): The image to get the channel dimension axis of. input_data_format (`ChannelDimension` or `str`, *optional*): The channel dimension format of the image. If `None`, will infer the channel dimension from the image. Returns: The channel dimension axis of the image. Nr:r Unsupported data format: )rLrr r=r!r>)r3rMr"r"r#get_channel_dimension_axiss    rO channel_dimcCsZ|durt|}|tjkr|jd|jdfS|tjkr&|jd|jdfStd|)a Returns the (height, width) dimensions of the image. Args: image (`np.ndarray`): The image to get the dimensions of. channel_dim (`ChannelDimension`, *optional*): Which dimension the channel dimension is in. If `None`, will infer the channel dimension from the image. Returns: A tuple of the image's height and width. NrN)rLrr rIr!r>)r3rPr"r"r#get_image_sizes   rT annotationcCsVt|tr)d|vr)d|vr)t|dttfr)t|ddks't|ddtr)dSdS)Nimage_id annotationsrTFr%dictr.r/lenrUr"r"r#"is_valid_annotation_coco_detections"r\cCs^t|tr-d|vr-d|vr-d|vr-t|dttfr-t|ddks+t|ddtr-dSdS)NrV segments_info file_namerTFrXr[r"r"r#!is_valid_annotation_coco_panoptics"r_rWcCtdd|DS)Ncs|]}t|VqdSr$)r\.0annr"r"r# z3valid_coco_detection_annotations..allrWr"r"r# valid_coco_detection_annotationsrjcCr`)Ncsrar$)r_rbr"r"r#rerfz2valid_coco_panoptic_annotations..rgrir"r"r#valid_coco_panoptic_annotationsrkrltimeoutc Csttdgt|trd|ds|dr#tjtj |d|dj }nOt j |r0tj|}nB|dr<|dd}ztj|dd }tjt|}Wn$tyc}z td |d |d }~wwt|tjjrn|}ntd tj|}|d}|S)a3 Loads `image` to a PIL Image. Args: image (`str` or `PIL.Image.Image`): The image to convert to the PIL Image format. timeout (`float`, *optional*): The timeout value in seconds for the URL request. Returns: `PIL.Image.Image`: A PIL Image. visionzhttp://zhttps://T)streamrmz data:image/,r )validatezIncorrect image source. Must be a valid URL starting with `http://` or `https://`, a valid path to an image file, or a base64 encoded string. Got z. Failed with NzuIncorrect format used for image. Should be an url linking to an image, a base64 string, a local path, or a PIL image.RGB)r load_imager%str startswithr&r'openrequestsgetrawospathisfilesplitbase64 b64decoder Exceptionr>ImageOpsexif_transposeconvert)r3rmb64er"r"r#rss2     rsc@seZdZdZddZdddZddZd ejd e e e fd ejfd d Z d ddZ ddZd!ddZd"ddZddZddZd#ddZdS)$ImageFeatureExtractionMixinzD Mixin that contain utilities for preparing image features. cCs8t|tjjtjfst|stdt|ddSdS)Nz Got type zS which is not supported, only `PIL.Image.Image`, `np.array` and `torch.Tensor` are.)r%r&r'r+r,rr>r?selfr3r"r"r#_ensure_format_supportedIs z4ImageFeatureExtractionMixin._ensure_format_supportedNcCs||t|r |}t|tjrE|dur t|jdtj}|jdkr3|j ddvr3| ddd}|r9|d}| tj }t j|S|S)a" Converts `image` to a PIL Image. Optionally rescales it and puts the channel dimension back as the last axis if needed. Args: image (`PIL.Image.Image` or `numpy.ndarray` or `torch.Tensor`): The image to convert to the PIL Image format. rescale (`bool`, *optional*): Whether or not to apply the scaling factor (to make pixel values integers between 0 and 255). Will default to `True` if the image type is a floating type, `False` otherwise. Nrr:rEr rF)rrnumpyr%r+r,flatfloatingr=rI transposeastyper6r&r' fromarray)rr3rescaler"r"r# to_pil_imagePs    z(ImageFeatureExtractionMixin.to_pil_imagecCs&||t|tjjs|S|dS)z Converts `PIL.Image.Image` to RGB format. Args: image (`PIL.Image.Image`): The image to convert. rr)rr%r&r'rrr"r"r# convert_rgbns  z'ImageFeatureExtractionMixin.convert_rgbr3scaler4cCs||||S)z7 Rescale a numpy image by scale amount )r)rr3rr"r"r#r|s z#ImageFeatureExtractionMixin.rescaleTcCs||t|tjjrt|}t|r|}|dur&t|jdtj n|}|r4| | tj d}|rB|j dkrB|ddd}|S)a Converts `image` to a numpy array. Optionally rescales it and puts the channel dimension as the first dimension. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image to convert to a NumPy array. rescale (`bool`, *optional*): Whether or not to apply the scaling factor (to make pixel values floats between 0. and 1.). Will default to `True` if the image is a PIL Image or an array/tensor of integers, `False` otherwise. channel_first (`bool`, *optional*, defaults to `True`): Whether or not to permute the dimensions of the image to put the channel dimension first. Nrp?r:rFr )rr%r&r'r+rBrrrintegerrrfloat32r=r)rr3r channel_firstr"r"r#rCs  z*ImageFeatureExtractionMixin.to_numpy_arraycCsD||t|tjjr|St|r|d}|Stj|dd}|S)z Expands 2-dimensional `image` to 3 dimensions. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image to expand. r)axis)rr%r&r'r unsqueezer+ expand_dimsrr"r"r#rs  z'ImageFeatureExtractionMixin.expand_dimsFcCs8||t|tjjr|j|dd}n|r3t|tjr'||tj d}n t |r3|| d}t|tjrXt|tjsHt ||j }t|tjsWt ||j }nt |rvddl}t||jsk||}t||jsv||}|jdkr|jddvr||ddddf|ddddfS|||S)a  Normalizes `image` with `mean` and `std`. Note that this will trigger a conversion of `image` to a NumPy array if it's a PIL Image. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image to normalize. mean (`List[float]` or `np.ndarray` or `torch.Tensor`): The mean (per channel) to use for normalization. std (`List[float]` or `np.ndarray` or `torch.Tensor`): The standard deviation (per channel) to use for normalization. rescale (`bool`, *optional*, defaults to `False`): Whether or not to rescale the image to be between 0 and 1. If a PIL image is provided, scaling will happen automatically. T)rrrNr:rE)rr%r&r'rCr+r,rrrrfloatrBr5torchTensortensorr=rI)rr3meanstdrrr"r"r# normalizes.         ( z%ImageFeatureExtractionMixin.normalizec CsJ|dur|ntj}||t|tjjs||}t|tr#t|}t|t s.t |dkr|rBt|t r9||fn|d|df}n\|j \}}||krO||fn||f\}} t|t r\|n|d} || krf|S| t | | |} } |dur|| krt d|d|| |krt || | |} } ||kr| | fn| | f}|j ||dS)a Resizes `image`. Enforces conversion of input to PIL.Image. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image to resize. size (`int` or `Tuple[int, int]`): The size to use for resizing the image. If `size` is a sequence like (h, w), output size will be matched to this. If `size` is an int and `default_to_square` is `True`, then image will be resized to (size, size). If `size` is an int and `default_to_square` is `False`, then smaller edge of the image will be matched to this number. i.e, if height > width, then image will be rescaled to (size * height / width, size). resample (`int`, *optional*, defaults to `PILImageResampling.BILINEAR`): The filter to user for resampling. default_to_square (`bool`, *optional*, defaults to `True`): How to convert `size` when it is a single int. If set to `True`, the `size` will be converted to a square (`size`,`size`). If set to `False`, will replicate [`torchvision.transforms.Resize`](https://pytorch.org/vision/stable/transforms.html#torchvision.transforms.Resize) with support for resizing only the smallest edge and providing an optional `max_size`. max_size (`int`, *optional*, defaults to `None`): The maximum allowed for the longer edge of the resized image: if the longer edge of the image is greater than `max_size` after being resized according to `size`, then the image is resized again so that the longer edge is equal to `max_size`. As a result, `size` might be overruled, i.e the smaller edge may be shorter than `size`. Only used if `default_to_square` is `False`. Returns: image: A resized `PIL.Image.Image`. Nr rz max_size = zN must be strictly greater than the requested size for the smaller edge size = )resample)PILImageResamplingBILINEARrr%r&r'rr.r/rHrZsizer>resize) rr3rrdefault_to_squaremax_sizewidthheightshortlongrequested_new_short new_shortnew_longr"r"r#rs4   $ z"ImageFeatureExtractionMixin.resizecCs||t|ts||f}t|st|tjr8|jdkr"||}|jddvr0|jddn|jdd}n |j d|j df}|d|dd}||d}|d|dd}||d}t|t j j rr| ||||fS|jddvr{dnd}|st|tjr| ddd}t|r|ddd}|dkr||dkr|dkr||dkr|d||||fS|jdd t|d|dt|d|df} t|tjrtj|| d } n t|r|| } | d |dd} | |d} | d |dd} | |d}|| d| | | |f<|| 7}|| 7}|| 7}|| 7}| dtd|t| jd |td|t| jd |f} | S) a Crops `image` to the given size using a center crop. Note that if the image is too small to be cropped to the size given, it will be padded (so the returned result has the size asked). Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor` of shape (n_channels, height, width) or (height, width, n_channels)): The image to resize. size (`int` or `Tuple[int, int]`): The size to which crop the image. Returns: new_image: A center cropped `PIL.Image.Image` or `np.ndarray` or `torch.Tensor` of shape: (n_channels, height, width). rFrrEr NTF.rQ)rIrR)rr%r/rr+r,r=rrIrr&r'croprpermuter8 zeros_like new_zerosr7)rr3r image_shapetopbottomleftrightr new_shape new_imagetop_pad bottom_padleft_pad right_padr"r"r# center_crop(sP    ,   (2    4z'ImageFeatureExtractionMixin.center_cropcCs>||t|tjjr||}|dddddddfS)a Flips the channel order of `image` from RGB to BGR, or vice versa. Note that this will trigger a conversion of `image` to a NumPy array if it's a PIL Image. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image whose color channels to flip. If `np.ndarray` or `torch.Tensor`, the channel dimension should be first. NrR)rr%r&r'rCrr"r"r#flip_channel_orderss  z.ImageFeatureExtractionMixin.flip_channel_orderrcCsL|dur|ntjj}||t|tjjs||}|j||||||dS)a Returns a rotated copy of `image`. This method returns a copy of `image`, rotated the given number of degrees counter clockwise around its centre. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image to rotate. If `np.ndarray` or `torch.Tensor`, will be converted to `PIL.Image.Image` before rotating. Returns: image: A rotated `PIL.Image.Image`. N)rexpandcenter translate fillcolor)r&r'NEARESTrr%rrotate)rr3anglerrrrrr"r"r#rs   z"ImageFeatureExtractionMixin.rotater$)NT)F)NTN)NrNNN)rrr__doc__rrrr+r,r rrHrrCrrrrrrr"r"r"r#rDs "   .CKr)r:r$)Br~rziortypingrrrrrrr rr+rw packagingr utilsr r rrrrrrutils.constantsrrrrrr PIL.Imager& PIL.ImageOpsparse __version__ base_versionr' Resamplingrrr, ImageInputrr*r-r0r2boolr9rHrArCrLrtrOrTr\r_rjrlrrsrr"r"r"r#sh $ (      '  ! """&&$-