tj>tddlZddlZddlZddlmcmcmZddl m Z ddl m Z ddl mZmZmZmZddlmZmZmZmZmZddlmZmZmZmZddlmZd d lm Z d d l!m"Z"erdd l#m$Z%d Z&dZ'dZ(eGdde Z)eedGdde"Z*ddgZ+dS)N)TorchvisionBackend) BatchFeature)group_images_by_shapepadreorder_imagesto_channel_dimension_format)OPENAI_CLIP_MEANOPENAI_CLIP_STDChannelDimensionPILImageResamplingSizeDict) TensorTypeauto_docstringis_scipy_availablerequires_backends)requires)OwlViTImageProcessor)OwlViTImageProcessorPil)ndimagec*t|}t|}|j}||jkr&|d||jz zz }t j||}n:||jdz kr||jdfz}n||jkrt d||fS)a%Validate resize output shape according to input image. Args: image (`np.ndarray`): Image to be resized. output_shape (`iterable`): Size of the generated output image `(rows, cols[, ...][, dim])`. If `dim` is not provided, the number of channels is preserved. Returns image (`np.ndarray`): The input image, but with additional singleton dimensions appended in the case where `len(output_shape) > input.ndim`. output_shape (`Tuple`): The output shape converted to tuple. Raises ------ ValueError: If output_shape length is smaller than the image number of dimensions. Notes ----- The input image is reshaped if its number of dimensions is not equal to output_shape_length. rzIoutput_shape length cannot be smaller than the image number of dimensions)tuplelenshapendimnpreshape ValueError)image output_shape output_ndim input_shapes e/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/owlv2/modular_owlv2.py_preprocess_resize_output_shaper(3s.&&Ll##K+KUZt{UZ788  5+..  Q & &#u{2&88 uz ! !deee , ctj|}tj|r$tj}tj}||}ntj}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. )r minisnannanminnanmaxmaxclip) input_image output_imagemin_valmin_funcmax_funcmax_vals r'_clip_warp_outputr7Zszf[!!G x99(;''66h{##G7<'::L r)ct|ttfr=tjd|D}tjd|D}nBt|tjr|d\}}ntdtj||}tj ||||gd}| d |j }||z}|S)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. cg|] }|d S)r.0is r' z _scale_boxes..s$@$@$@aQqT$@$@$@r)cg|] }|d Srr:r;s r'r>z _scale_boxes..s#?#?#?QAaD#?#?#?r)rz4`target_sizes` must be a list, tuple or torch.Tensor)dim) isinstancelistrtorchtensorTensorunbind TypeErrorr/stack unsqueezetodevice)boxes target_sizes image_height image_widthmax_size scale_factors r' _scale_boxesrRxs,u ..P|$@$@<$@$@$@AA l#?#?,#?#?#?@@ L%, / /P$0$7$7$:$:! kkNOOOy{33H;(HhGQOOOL))!,,// ==L L E Lr)cBeZdZejZeZeZ dddZ dZ dZ dZ dZdZdZdZdddd ed dfd Z ddedd edzd ed edfd Z ddddeded dfdZdeddededddededededeeezdzdeeezdzd edzdeezdzd efdZdS)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) selfr[r\rWrXsize pad_bottom pad_rightpadding padded_images r' _pad_imageszOwlv2ImageProcessor._pad_imagessc RSS) 65!!F] 5L aJ/wvw^DDD r)disable_groupingc t||\}}i}|D]!\}} || |} | ||<"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. rj)r\)ritemsrir) rcr[rjr\kwargsgrouped_imagesgrouped_images_indexprocessed_images_groupedrstacked_imagesprocessed_imagess r'rzOwlv2ImageProcessor.pads0EV^n/o/o/o,,#% %3%9%9%;%; = = !E>!---.N/= $U + +)*BDXYYr)r#rd anti_aliasingc n|j|jf}|j}tj|dd|jtj||jz }|r$||dz dz d}ntj|tj |z}tj |dkrtdtj |dk|dkzrtj dtj |dkr|} nidtjd|zzdz} t!j|| d| df| } 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)rd antialias)rWrXrrCrDrJrKclamp atleast_1d ones_likeanyr"warningswarnceilintrb gaussian_blurtolistrresize) rcr#rdrtanti_aliasing_sigmarnr$r&factorsfiltered kernel_sizess r'rzOwlv2ImageProcessor.resizes, TZ0 k ,{122//225<@@5!77 NE:yN/= $U + +)*BDXYY  q#xx(8_oxpp /D /?0 0 0 ,,"$%3%9%9%;%; > > !E> > $ .tV^ _ _ 0=&u-'(>@TUU0E^fv/w/w/w,,#% %3%9%9%;%; = = !E>!77~|ZQZN/= $U + +)*BDXYY.2B!CQ_````r)rZTN)__name__ __module__ __qualname__r BILINEARrr rr rrdrrrrr crop_sizedo_center_cropfloatrirBboolrrrstrrrrr:r)r'rTrTs!*H!JIC ( (DNIJL FIN  . % R`    "!$   ^$ +   n      8# 4O4O4O4O 4O 4O4O4O4Ol7a^$7a7a 7a . 7a  7a7a7a7aDK'$.7a4;&-7a+7aj(4/7a 7a7a7a7a7a7ar)rT)rC)backendsc@eZdZejZeZeZ dddZ dZ dZ dZ dZdZdZdZddejded ejfd Z ddejd eeefd ed ejfd Zdeejded edddededededeeezdzdeeezdzdeezdzd efdZdS)Owlv2ImageProcessorPilrUrVrYTNrZr#r\r]c|jdd\}}t||}||z }||z }t|d|fd|ff|}|S)r_r`Nr)r#rgconstant_values)rr/r)rcr#r\rWrXrdrerfs r'rzOwlv2ImageProcessorPil.padIsj BCC( 65!!F] 5L _q)n5*     r)rdrtc 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 |dkrtd t j |dk|dkzrtjd tj||| | } n|} d |D} tj| | | | | d }t%||}t|tj}|S)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. scipyrWrXmirrorrrNrrvrw)cvalmodecg|]}d|z Srr:)r<fs r'r>z1Owlv2ImageProcessorPil.resize..s///!A///r)T)orderrr grid_mode)rr r LASTr(rr dividemaximumr{r|r}r"r~rndigaussian_filterzoomr7FIRST)rcr#rdrtrrnr$r&rndi_moderrr zoom_factorsouts r'rzOwlv2ImageProcessorPil.resizeXs, $(((XW 6 +E3C3HII=e\RR|k )K66  "*&(jWq[A4E&F&F##&(m4G&H&H2rso 222222222222;;;;;;222222  +*****AAAAAAHHHHHH%$$$$$$$$$N<>_a_a_a_a_a._a_a_aD :oaoaoaoaoa4oaoaoad !": ;r)