i>HddlZddlZddlZddlmcmcmZddl m Z ddl m Z ddl mZmZmZmZddlmZmZmZmZmZddlmZmZmZmZddlmZd d lm Z d d l!m"Z"erdd l#m$Z%d Z&dZ'dZ(eGdde Z)eedGdde"Z*ddgZ+y)N)TorchvisionBackend) BatchFeature)group_images_by_shapepadreorder_imagesto_channel_dimension_format)OPENAI_CLIP_MEANOPENAI_CLIP_STDChannelDimensionPILImageResamplingSizeDict) TensorTypeauto_docstringis_scipy_availablerequires_backends)requires)OwlViTImageProcessor)OwlViTImageProcessorPil)ndimagecPt|}t|}|j}||jkDr/|d||jz zz }t j ||}||fS||jdz k(r||jdfz}||fS||jkr t d||fS)a%Validate resize output shape according to input image. Args: image (`np.ndarray`): Image to be resized. output_shape (`iterable`): Size of the generated output image `(rows, cols[, ...][, dim])`. If `dim` is not provided, the number of channels is preserved. Returns image (`np.ndarray`): The input image, but with additional singleton dimensions appended in the case where `len(output_shape) > input.ndim`. output_shape (`Tuple`): The output shape converted to tuple. Raises ------ ValueError: If output_shape length is smaller than the image number of dimensions. Notes ----- The input image is reshaped if its number of dimensions is not equal to output_shape_length. )rzIoutput_shape length cannot be smaller than the image number of dimensions)tuplelenshapendimnpreshape ValueError)image output_shape output_ndim input_shapes x/var/www/vps2.regionflexible.com/Desarrollo/venv/lib/python3.12/site-packages/transformers/models/owlv2/modular_owlv2.py_preprocess_resize_output_shaper'3s.&Ll#K++KUZZt{UZZ788  5+. ,   Q &#u{{2&88  ,  uzz !dee , c*tj|}tj|r)tj}tj}||}n tj}tj }||}tj |||}|S)aClip output image to range of values of input image. Note that this function modifies the values of *output_image* in-place. Taken from: https://github.com/scikit-image/scikit-image/blob/b4b521d6f0a105aabeaa31699949f78453ca3511/skimage/transform/_warps.py#L640. Args: input_image : ndarray Input image. output_image : ndarray Output image, which is modified in-place. )rminisnannanminnanmaxmaxclip) input_image output_imagemin_valmin_funcmax_funcmax_vals r&_clip_warp_outputr6Zsnff[!G xx9999;'6666{#G77<':L r(ct|ttfrMtj|Dcgc]}|d c}}tj|Dcgc]}|d c}}n:t|tj r|j d\}}n tdtj||}tj||||gd}|jdj|j}||z}|Scc}wcc}w)a Scale batch of bounding boxes to the target sizes. Args: boxes (`torch.Tensor` of shape `(batch_size, num_boxes, 4)`): Bounding boxes to scale. Each box is expected to be in (x1, y1, x2, y2) format. target_sizes (`list[tuple[int, int]]` or `torch.Tensor` of shape `(batch_size, 2)`): Target sizes to scale the boxes to. Each target size is expected to be in (height, width) format. Returns: `torch.Tensor` of shape `(batch_size, num_boxes, 4)`: Scaled bounding boxes. rrz4`target_sizes` must be a list, tuple or torch.Tensor)dim) isinstancelistrtorchtensorTensorunbind TypeErrorr.stack unsqueezetodevice)boxes target_sizesi image_height image_widthmax_size scale_factors r& _scale_boxesrKxs,u .||<$@aQqT$@A ll,#?QAaD#?@ L%,, /$0$7$7$:! kNOOyy{3H;;(HhGQOL))!,// =L L E L%A#?s D Dc,eZdZejZeZeZ dddZ dZ dZ dZ dZdZdZdZdddded dfd Z ddedd edzded edfd Z dd ddeded dfdZdeddededddededededeeezdzdeeezdzd edzdeezdzd efdZy)Owlv2ImageProcessorheightwidthp?TNimagesz torch.Tensorconstant_valuereturnc|jdd\}}t||}||z }||z }dd||f}tj|||} | S)< Pad an image with zeros to the given size. Nr)fill)rr.tvFr) selfrSrTrPrQsize pad_bottom pad_rightpadding padded_images r& _pad_imageszOwlv2ImageProcessor._pad_imagess\ RS) 65!F] 5L aJ/wwvw^D r(disable_groupingc t||\}}i}|jD]\}} |j| |} | ||<t||} | S)z Unlike the Base class `self.pad` where all images are padded to the maximum image size, Owlv2 pads an image to square. rb)rT)ritemsrar) r[rSrbrTkwargsgrouped_imagesgrouped_images_indexprocessed_images_groupedrstacked_imagesprocessed_imagess r&rzOwlv2ImageProcessor.padsz0EV^n/o,,#% %3%9%9%; = !E>!---.N/= $U +  =**BDXYr(r"r\ anti_aliasingc h|j|jf}|j}tj|ddj |j tj|j |j z }|r||dz dz jd}ntj|tj|z}tj|dkr tdtj|dkD|dkzrtjdtj|dk(r|} n]dtjd|zjzdz} t!j"|| d| df|j% } n|} t'j(| |d S) az Resize an image as per the original implementation. Args: image (`Tensor`): Image to resize. size (`dict[str, int]`): Dictionary containing the height and width to resize the image to. anti_aliasing (`bool`, *optional*, defaults to `True`): Whether to apply anti-aliasing when downsampling the image. anti_aliasing_sigma (`float`, *optional*, defaults to `None`): Standard deviation for Gaussian kernel when downsampling the image. If `None`, it will be calculated automatically. rNrr)r*FAnti-aliasing standard deviation must be greater than or equal to zeroWAnti-aliasing standard deviation greater than zero but not down-sampling along all axesr)sigmaF)r\ antialias)rPrQrr;r<rBrCclamp atleast_1d ones_likeanyr!warningswarnceilintrZ gaussian_blurtolistrresize) r[r"r\rlanti_aliasing_sigmarfr#r%factorsfiltered kernel_sizess r&r|zOwlv2ImageProcessor.resizes~, TZZ0 kk ,,{12/225<<@5<!77 NE:yN/= $U +  =**BDXY #xx(8_oxp /D /?0 ,,"$%3%9%9%; > !E> $ .tV^ _ 0=&u- >((>@TU0E^fv/w,,#% %3%9%9%; = !E>!77~|ZQZN/= $U +  =**BDXY.2B!CQ_``r(rTN)__name__ __module__ __qualname__r BILINEARrr rr rr\rrrrr crop_sizedo_center_cropfloatrar:boolrrr|strrrrr(r&rMrMs!**H!JIC (DNIJL FIN . % R` "!$  ^$ +   n   8# 4O4O4O 4O 4Ol7a^$7a7a 7a . 7a  7a7a7a7aDK'$.7a4;&-7a+7aj(4/7a 7ar(rM)r;)backendscbeZdZejZeZeZ dddZ dZ dZ dZ dZdZdZdZddej&dedej&fd Z ddej&d eeefd edej&fd Zd eej&ded edddededededeeezdzdeeezdzdeezdzdefdZy)Owlv2ImageProcessorPilrNrOrRTNr"rTrUc~|jdd\}}t||}||z }||z }t|d|fd|ff|}|S)rWrXNr)r"r_constant_values)rr.r)r[r"rTrPrQr\r]r^s r&rzOwlv2ImageProcessorPil.padIs] BC( 65!F] 5L _q)n5*   r(r\rlc t|d|d|df}t|tj}t ||\}}|j }t j||}d} d} d} |r|t jd|dz dz }nt j|t j|z}t j|dkr tdt j|dkD|dkzrtjd tj ||| | } n|} |D cgc]} d| z  }} tj"| || | | d }t%||}t|tj&}|Scc} w) a~ Resize an image as per the original implementation. Args: image (`np.ndarray`): Image to resize. size (`dict[str, int]`): Dictionary containing the height and width to resize the image to. anti_aliasing (`bool`, *optional*, defaults to `True`): Whether to apply anti-aliasing when downsampling the image. anti_aliasing_sigma (`float`, *optional*, defaults to `None`): Standard deviation for Gaussian kernel when downsampling the image. If `None`, it will be calculated automatically. scipyrPrQmirrorrrrrnro)cvalmodeT)orderrr grid_mode)rr r LASTr'rrdividemaximumrsrtrur!rvrwndigaussian_filterzoomr6FIRST)r[r"r\rlr}rfr#r%r~ndi_moderrrf zoom_factorsouts r&r|zOwlv2ImageProcessorPil.resizeXsq, $(XW 6 +E3C3H3HI=e\R|kk ))K6 "*&(jjWq[A4E&F#&(mm4G&H2<rs 22;2  +AH$$N<>_a._a_aD :oa4oaoad !": ;r(