U 9%ed@sddlZddlZddlmZddlmZmZmZmZm Z m Z m Z ddl Z ddlZddlmZddlmZmZmZmZmZmZmZmZddlmZmZmZmZmZm Z erddl!Z"ddl#Z"e$e$e"j%j&e$dkre"j'j(Z)ne"j'Z)ererddl*Z*e d e j+d ed ee j+ed fZ,Gd d d eZ-d dZ.ddZ/ddZ0ddZ1e j+e2dddZ3d8e4ee,dddZ5e j+dddZ6d9e j+e e e4e e4dffe-d d!d"Z7d:e j+e e e-e8fe4d#d$d%Z9d;e j+e-e e4e4fd&d'd(Z:ee8e ee ffe2d)d*d+Z;ee8e ee ffe2d)d,d-Zdsz3valid_coco_detection_annotations..allrQr r r! valid_coco_detection_annotationssr`cCstdd|DS)Ncss|]}t|VqdSr")rXrYr r r!r\sz2valid_coco_panoptic_annotations..r]r_r r r!valid_coco_panoptic_annotationssra)r0timeoutr1c Cs ttdgt|tr|ds*|drFtjtj |d|dj }qt j |r`tj|}q|drx|dd}z"tj|dd }tjt|}Wqtk r}ztd |d |W5d }~XYqXnt|tjjr|}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. Zvisionzhttp://zhttps://T)streamrbz 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$r%openrequestsgetrawospathisfilesplitbase64 b64decoder Exceptionr;ZImageOpsZexif_transposeconvert)r0rbZb64er r r!rgs.      rgc@seZdZdZddZdddZddZeje e e fejd d d Z dd dZ ddZdddZd ddZddZddZd!ddZdS)"ImageFeatureExtractionMixinzD Mixin that contain utilities for preparing image features. cCs4t|tjjtjfs0t|s0tdt|ddS)Nz Got type zS which is not supported, only `PIL.Image.Image`, `np.array` and `torch.Tensor` are.)r#r$r%r)r*rr;r<selfr0r r r!_ensure_format_supportedKsz4ImageFeatureExtractionMixin._ensure_format_supportedNcCs||t|r|}t|tjr|dkr@t|jdtj}|jdkrf|j ddkrf| ddd}|rr|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. Nrr7rAr rB)rzrnumpyr#r)r*flatZfloatingr:rE transposeastyper3r$r%Z fromarray)ryr0rescaler r r! to_pil_imageRs    z(ImageFeatureExtractionMixin.to_pil_imagecCs&||t|tjjs|S|dS)z Converts `PIL.Image.Image` to RGB format. Args: image (`PIL.Image.Image`): The image to convert. rf)rzr#r$r%rurxr r r! convert_rgbps z'ImageFeatureExtractionMixin.convert_rgb)r0scaler1cCs||||S)z7 Rescale a numpy image by scale amount )rz)ryr0rr r r!r~s z#ImageFeatureExtractionMixin.rescaleTcCs||t|tjjr"t|}t|r2|}|dkrLt|jdtj n|}|rh| | tj d}|r|j dkr|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?r7rBr )rzr#r$r%r)r>rr|r}integerrrfloat32r:r~)ryr0r channel_firstr r r!r?s  z*ImageFeatureExtractionMixin.to_numpy_arraycCsB||t|tjjr|St|r0|d}ntj|dd}|S)z Expands 2-dimensional `image` to 3 dimensions. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image to expand. r)Zaxis)rzr#r$r%rZ unsqueezer) expand_dimsrxr r r!rs  z'ImageFeatureExtractionMixin.expand_dimsFcCs@||t|tjjr(|j|dd}n>|rft|tjrN||tj d}nt |rf|| d}t|tjrt|tjst ||j }t|tjst ||j }nr2torchZTensorZtensorr:rE)ryr0ZmeanZstdrrr r r! normalizes,         (z%ImageFeatureExtractionMixin.normalizec CsT|dk r |ntj}||t|tjjs4||}t|trFt|}t|t s^t |dkrF|rt|t rt||fn|d|df}n|j \}}||kr||fn||f\}} t|t r|n|d} || kr|S| t | | |} } |dk r,|| kr t d|d|| |kr,t || | |} } ||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)PILImageResamplingZBILINEARrzr#r$r%rr,r-rDrTsizer;resize) ryr0rrZdefault_to_squaremax_sizewidthheightshortlongZrequested_new_shortZ new_shortZnew_longr r r!rs0   $    z"ImageFeatureExtractionMixin.resizecCs||t|ts||f}t|s0t|tjrp|jdkrD||}|jddkr`|jddn |jdd}n|j d|j df}|d|dd}||d}|d|dd}||d}t|t j j r| ||||fS|jddkrdnd}|s4t|tjr| ddd}t|r4|ddd}|dkrz||dkrz|dkrz||dkrz|d||||fS|jdd t|d|dt|d|df} t|tjrtj|| d } nt|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). rBrrAr NTF.rK)rErL)rzr#r-rr)r*r:rrErr$r%cropr~Zpermuter5Z zeros_likeZ new_zerosr4)ryr0rZ image_shapetopbottomleftrightrZ new_shapeZ new_imageZtop_padZ bottom_padZleft_padZ right_padr r r! center_crop*sZ    ,   02    z'ImageFeatureExtractionMixin.center_cropcCs>||t|tjjr"||}|dddddddfS)a Flips the channel order of `image` from RGB to BGR, or vice versa. Note that this will trigger a conversion of `image` to a NumPy array if it's a PIL Image. Args: image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`): The image whose color channels to flip. If `np.ndarray` or `torch.Tensor`, the channel dimension should be first. NrL)rzr#r$r%r?rxr r r!flip_channel_orderus  z.ImageFeatureExtractionMixin.flip_channel_orderrcCsL|dk r |ntjj}||t|tjjs6||}|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$r%ZNEARESTrzr#rrotate)ryr0Zanglerrrrrr r r!rs   z"ImageFeatureExtractionMixin.rotate)N)NT)F)NTN)NrNNN)rrr__doc__rzrrr)r*r rrDrr?rrrrrrr r r r!rwFs   . CKrw)r7)N)N)N)N)Brrrniortypingrrrrrrr r|r)rk packagingr utilsr r rrrrrrZutils.constantsrrrrrrZ PIL.Imager$Z PIL.ImageOpsparse __version__ base_versionr%Z Resamplingrrr*Z ImageInputrr(r+r.r/boolr6rDr=r?rFrhrIrNrVrXr`rarrgrwr r r r!s^ $ (     '  !   $$ -