o i. @sddlZddlmZddlmZmZmZddlZddlm Z ddl m Z m Z ddl mZmZddlmZmZmZmZdd lmZmZmZd d lmZerRdd lmZdd lmZdddee ee fde!de def ddZ"Gddde Z#eGdddeZ$dgZ%dS)N)Iterable) TYPE_CHECKINGOptionalUnion) BatchFeature)BaseImageProcessorFastDefaultFastImageProcessorKwargs)group_images_by_shapereorder_images)IMAGENET_STANDARD_MEANIMAGENET_STANDARD_STDPILImageResamplingSizeDict) TensorTypeauto_docstringrequires_backends)BeitImageProcessorFast)DepthEstimatorOutput) functional input_image torch.Tensor output_sizekeep_aspect_ratiomultiplereturnc Csddd}|jdd\}}|\}}||} ||} |r-td| td| kr+| } n| } || ||d} || ||d} t| | dS) NrcSsPt|||}|dur||krt|||}||kr&t|||}|SN)roundmathfloorceil)valrmin_valmax_valxr&W/home/ubuntu/.local/lib/python3.10/site-packages/transformers/models/dpt/modular_dpt.pyconstrain_to_multiple_of4s z>get_resize_output_image_size..constrain_to_multiple_of)rheightwidth)rN)shapeabsr) rrrrr( input_height input_width output_height output_width scale_height scale_width new_height new_widthr&r&r'get_resize_output_image_size.s   r8c@sBeZdZUdZeeed<eeed<eeed<eeed<dS)DPTFastImageProcessorKwargsa ensure_multiple_of (`int`, *optional*, defaults to 1): If `do_resize` is `True`, the image is resized to a size that is a multiple of this value. Can be overridden by `ensure_multiple_of` in `preprocess`. size_divisor (`int`, *optional*): If `do_pad` is `True`, pads the image dimensions to be divisible by this value. This was introduced in the DINOv2 paper, which uses the model in combination with DPT. keep_aspect_ratio (`bool`, *optional*, defaults to `False`): If `True`, the image is resized to the largest possible size such that the aspect ratio is preserved. Can be overridden by `keep_aspect_ratio` in `preprocess`. do_reduce_labels (`bool`, *optional*, defaults to `self.do_reduce_labels`): Whether or not to reduce all label values of segmentation maps by 1. Usually used for datasets where 0 is used for background, and background itself is not included in all classes of a dataset (e.g. ADE20k). The background label will be replaced by 255. ensure_multiple_of size_divisorrdo_reduce_labelsN)__name__ __module__ __qualname____doc__rint__annotations__boolr&r&r&r'r9Us    r9c'@s|eZdZejZeZeZ dddZ dZ dZ dZ dZdZdZdZdZdZdZdZeZ    d*dd d ed ed d edeededd fddZ d+dd dedd fddZded deded ed ed dedededededeeeeefdeeeeefdedeed edeed!eed"eee e!fde"f&d#d$Z# d,d%d&d'eee!ee$eefdfdee%e e!ffd(d)Z&dS)-DPTImageProcessorFastir+TFgp?r*Nimagersize interpolationzF.InterpolationMode antialiasr:rrcCsJ|jr|jstd|t||j|jf||d}tj|||||dS)a< Resize an image to `(size["height"], size["width"])`. Args: image (`torch.Tensor`): Image to resize. size (`SizeDict`): Dictionary in the format `{"height": int, "width": int}` specifying the size of the output image. interpolation (`InterpolationMode`, *optional*, defaults to `InterpolationMode.BILINEAR`): `InterpolationMode` filter to use when resizing the image e.g. `InterpolationMode.BICUBIC`. antialias (`bool`, *optional*, defaults to `True`): Whether to use antialiasing when resizing the image ensure_multiple_of (`int`, *optional*): If `do_resize` is `True`, the image is resized to a size that is a multiple of this value keep_aspect_ratio (`bool`, *optional*, defaults to `False`): If `True`, and `do_resize` is `True`, the image is resized to the largest possible size such that the aspect ratio is preserved. Returns: `torch.Tensor`: The resized image. zDThe size dictionary must contain the keys 'height' and 'width'. Got )rrr)rGrH)r,r- ValueErrorkeysr8rresize)selfrErFrGrHr:rrr&r&r'rKs   zDPTImageProcessorFast.resizer;c CsN|jdd\}}dd}|||\}}|||\}} ||| |f} t|| S)a Center pad a batch of images to be a multiple of `size_divisor`. Args: image (`torch.Tensor`): Image to pad. Can be a batch of images of dimensions (N, C, H, W) or a single image of dimensions (C, H, W). size_divisor (`int`): The width and height of the image will be padded to a multiple of this number. r)NcSs2t|||}||}|d}||}||fS)Nr)rr!)rFr;new_sizepad_size pad_size_leftpad_size_rightr&r&r'_get_pads z1DPTImageProcessorFast.pad_image.._get_pad)r.Fpad) rLrEr;r,r-rQpad_top pad_bottompad_left pad_rightpaddingr&r&r' pad_images   zDPTImageProcessorFast.pad_imageimagesr< do_resizedo_center_crop crop_size do_rescalerescale_factor do_normalize image_mean image_stddo_paddisable_groupingreturn_tensorsc Ks|r||}t||d\}}i}|D]\}}|r%|j||||| d}|||<qt||}t||d\}}i}|D]"\}}|rI|||}|rQ|||}|||| | | | }|||<q=t||}|rntj |ddn|}t d|idS)N)rd)rErFrGr:rr)dim pixel_values)data) reduce_labelr itemsrKr center_croprYrescale_and_normalizetorchstackr)rLrZr<r[rFrGr\r]r^r_r`rarbrr:rcr;rdrekwargsgrouped_imagesgrouped_images_indexresized_images_groupedr.stacked_imagesresized_imagesprocessed_images_groupedprocessed_imagesr&r&r' _preprocesss:        z!DPTImageProcessorFast._preprocessoutputsr target_sizescCst|d|j}|durt|t|krtdg}|dur%dgt|n|}t||D]"\}}|durGtjjj| d d|ddd }| d |iq,|S) a Converts the raw output of [`DepthEstimatorOutput`] into final depth predictions and depth PIL images. Only supports PyTorch. Args: outputs ([`DepthEstimatorOutput`]): Raw outputs of the model. target_sizes (`TensorType` or `List[Tuple[int, int]]`, *optional*): Tensor of shape `(batch_size, 2)` or list of tuples (`Tuple[int, int]`) containing the target size (height, width) of each image in the batch. If left to None, predictions will not be resized. Returns: `List[Dict[str, TensorType]]`: A list of dictionaries of tensors representing the processed depth predictions. rmNz]Make sure that you pass in as many target sizes as the batch dimension of the predicted depthrr*bicubicF)rFmode align_cornerspredicted_depth) rr}lenrIziprmnnr interpolate unsqueezesqueezeappend)rLrxryr}resultsdepth target_sizer&r&r'post_process_depth_estimations" z3DPTImageProcessorFast.post_process_depth_estimation)NTr*F)r*r)'r=r>r?rBICUBICresampler rar rbrFr[r^r`rcr_r:rr<r]r\r9 valid_kwargsrrrCrArKrYlistfloatrstrrrrwtupledictrr&r&r&r'rDls  -       >rD)&rcollections.abcrtypingrrrrmimage_processing_baserimage_processing_utils_fastrr image_transformsr r image_utilsr r rrutilsrrrbeit.image_processing_beit_fastrmodeling_outputsrtorchvision.transforms.v2rrRrArCr8r9rD__all__r&r&r&r's:      ' ?