§
    9²¸jäV  ã                   óÈ  — d Z ddlZddlZddlmZmZ ddlZddlZddl	m
Z ddlmZ ddlmZ ddlmZmZ dd	lmZmZmZmZmZmZ dd
lmZmZ ddlmZmZm Z m!Z!m"Z"  e ¦   «         rddl#m$Z$ ddl%m&Z&m'Z' erddl(m)Z)  e!j*        e+¦  «        Z, G d„ ded¬¦  «        Z-	 	 d1de.e/z  ej0        z  de1de1de2de2f
d„Z3de2dej0        dej0        dej0        fd „Z4d!„ Z5d"ej0        d#e.ej0        ej0        f         fd$„Z6d2d%ej0        d&ej0        d'e1d#ej0        fd(„Z7d%ej0        d)ej0        d*ej0        d+ej0        d#ej0        f
d,„Z8d-ej0        d#ej0        fd.„Z9e G d/„ d0e¦  «        ¦   «         Z:d0gZ;dS )3z"Image processor class for VitPose.é    N)ÚTYPE_CHECKINGÚUnion)Ú
functionalé   )ÚTorchvisionBackend)ÚBatchFeature)Úgroup_images_by_shapeÚreorder_images)ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STDÚChannelDimensionÚ
ImageInputÚPILImageResamplingÚSizeDict)ÚImagesKwargsÚUnpack)Ú
TensorTypeÚauto_docstringÚis_scipy_availableÚloggingÚrequires_backends)Úinv)Úaffine_transformÚgaussian_filteré   )ÚVitPoseEstimatorOutputc                   ó4   — e Zd ZU dZedz  ed<   edz  ed<   dS )ÚVitPoseImageProcessorKwargsaI  
    do_affine_transform (`bool`, *optional*):
        Whether to apply an affine transformation to the input images based on the bounding boxes.
    normalize_factor (`float`, *optional*, defaults to `200.0`):
        Width and height scale factor used for normalization when computing center and scale from bounding boxes.
    NÚdo_affine_transformÚnormalize_factor)Ú__name__Ú
__module__Ú__qualname__Ú__doc__ÚboolÚ__annotations__Úfloat© ó    ú„/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/transformers/models/vitpose/image_processing_vitpose.pyr   r   7   s>   € € € € € € ðð ð  ™Ð$Ð$Ñ$Ø˜d‘lÐ"Ð"Ñ"Ð"Ð"r)   r   F)Útotalç      i@ç      ô?ÚboxÚimage_widthÚimage_heightr    Úpadding_factorc                 ó(  — | dd…         \  }}}}||z  }	t          j        ||dz  z   ||dz  z   gt           j        ¬¦  «        }
||	|z  k    r	|dz  |	z  }n||	|z  k     r||	z  }t          j        ||z  ||z  gt           j        ¬¦  «        }||z  }|
|fS )a¥  
    Encodes a bounding box in COCO format into (center, scale).

    Args:
        box (`Tuple`, `List`, or `np.ndarray`):
            Bounding box in COCO format (top_left_x, top_left_y, width, height).
        image_width (`int`):
            Image width.
        image_height (`int`):
            Image height.
        normalize_factor (`float`):
            Width and height scale factor.
        padding_factor (`float`):
            Bounding box padding factor.

    Returns:
        tuple: A tuple containing center and scale.

        - `np.ndarray` [float32](2,): Center of the bbox (x, y).
        - `np.ndarray` [float32](2,): Scale of the bbox width & height.
    Né   ç      à?©Údtypeç      ð?)ÚnpÚarrayÚfloat32)r.   r/   r0   r    r1   Ú
top_left_xÚ
top_left_yÚwidthÚheightÚaspect_ratioÚcenterÚscales               r*   Úbox_to_center_and_scalerB   D   sÊ   € ð: -0°°°¬GÑ)€J�
˜E 6Ø Ñ-€LÝŒX�z E¨C¡KÑ/°¸fÀs¹lÑ1JÐKÕSUÔS]Ð^Ñ^Ô^€Fàˆ|˜fÑ$Ò$Ð$Ø˜‘˜|Ñ+ˆˆØ	� Ñ&Ò	&Ð	&Ø˜Ñ%ˆåŒH�eÐ.Ñ.°Ð9IÑ0IÐJÕRTÔR\Ð]Ñ]Ô]€EØ�NÑ"€Eà�5ˆ=Ðr)   ÚthetaÚ
size_inputÚsize_dstÚsize_targetc                 óÐ  — t          j        | ¦  «        } t          j        dt           j        ¬¦  «        }|d         |d         z  }|d         |d         z  }t	          j        | ¦  «        |z  |d<   t	          j        | ¦  «         |z  |d<   |d|d         z  t	          j        | ¦  «        z  d|d         z  t	          j        | ¦  «        z  z   d|d         z  z   z  |d	<   t	          j        | ¦  «        |z  |d
<   t	          j        | ¦  «        |z  |d<   |d|d         z  t	          j        | ¦  «        z  d|d         z  t	          j        | ¦  «        z  z
  d|d         z  z   z  |d<   |S )aÈ  
    Calculate the transformation matrix under the constraint of unbiased. Paper ref: Huang et al. The Devil is in the
    Details: Delving into Unbiased Data Processing for Human Pose Estimation (CVPR 2020).

    Source: https://github.com/open-mmlab/mmpose/blob/master/mmpose/core/post_processing/post_transforms.py

    Args:
        theta (`float`):
            Rotation angle in degrees.
        size_input (`np.ndarray`):
            Size of input image [width, height].
        size_dst (`np.ndarray`):
            Size of output image [width, height].
        size_target (`np.ndarray`):
            Size of ROI in input plane [w, h].

    Returns:
        `np.ndarray`: A matrix for transformation.
    )é   r   r5   r   r   ©r   r   ©r   r   g      à¿r4   ©r   rH   ©r   r   ©r   r   ©r   rH   )r8   Údeg2radÚzerosr:   ÚmathÚcosÚsin)rC   rD   rE   rF   ÚmatrixÚscale_xÚscale_ys          r*   Úget_warp_matrixrW   p   s[  € õ( ŒJ�uÑÔ€EÝŒX�f¥B¤JÐ/Ñ/Ô/€FØ�qŒk˜K¨œNÑ*€GØ�qŒk˜K¨œNÑ*€GÝ”8˜E‘?”? WÑ,€Fˆ4�LÝ”H˜U‘O”OÐ# gÑ-€Fˆ4�LØØˆz˜!Œ}Ñ�tœx¨™œÑ.°°zÀ!´}Ñ1DÅtÄxÐPUÁÄÑ1VÑVÐY\Ð_jÐklÔ_mÑYmÑmñ€Fˆ4�Lõ ”8˜E‘?”? WÑ,€Fˆ4�LÝ”8˜E‘?”? WÑ,€Fˆ4�LØØˆz˜!Œ}Ñ�tœx¨™œÑ.°°zÀ!´}Ñ1DÅtÄxÐPUÁÄÑ1VÑVÐY\Ð_jÐklÔ_mÑYmÑmñ€Fˆ4�Lð €Mr)   c                 óz  ‡ ‡‡— ˆ fd„t          ‰ j        d         ¦  «        D ¦   «         }t          j        |g d¢g¦  «        }t	          |¦  «        Š‰d         ‰d         ‰d         ‰d         ‰d         ‰d	         f\  ‰d<   ‰d<   ‰d<   ‰d<   ‰d	<   ‰d<   ˆˆfd
„|D ¦   «         }t          j        |d¬¦  «        }|S )a[  
    This function implements cv2.warpAffine function using affine_transform in scipy. See https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.affine_transform.html and https://docs.opencv.org/4.x/d4/d61/tutorial_warp_affine.html for more details.

    Note: the original implementation of cv2.warpAffine uses cv2.INTER_LINEAR.
    c                 ó$   •— g | ]}‰d |f         ‘ŒS ).r(   )Ú.0ÚiÚsrcs     €r*   ú
<listcomp>z%scipy_warp_affine.<locals>.<listcomp>›   s!   ø€ Ð:Ð:Ð: ��C˜�F”Ð:Ð:Ð:r)   éÿÿÿÿ)r   r   r   rM   rL   rJ   rI   rN   rK   c                 ó6   •— g | ]}t          |‰‰d ¬¦  «        ‘ŒS )r   )Úoutput_shapeÚorder)r   )rZ   ÚchannelÚM_invÚsizes     €€r*   r]   z%scipy_warp_affine.<locals>.<listcomp>ª   s,   ø€ ÐdÐdÐdÐPWÕ ¨¸TÈÐKÑKÔKÐdÐdÐdr)   ©Úaxis)ÚrangeÚshaper8   Úvstackr   Ústack)r\   ÚMrd   ÚchannelsÚM_scipyÚnew_srcrc   s   ` `   @r*   Úscipy_warp_affinero   •   sâ   øøø€ ð ;Ð:Ð:Ð:¥U¨3¬9°R¬=Ñ%9Ô%9Ð:Ñ:Ô:€Hõ Œi˜˜I˜I˜I˜Ñ'Ô'€Gå�‰LŒL€EàˆdŒØˆdŒØˆdŒØˆdŒØˆdŒØˆdŒðTÑP€Eˆ$�K��t‘˜e D™k¨5°©;¸¸d¹ÀUÈ4Á[ð eÐdÐdÐdÐdÐ[cÐdÑdÔd€GÝŒh�w RÐ(Ñ(Ô(€GØ€Nr)   ÚheatmapsÚreturnc                 ó²  — t          | t          j        ¦  «        st          d¦  «        ‚| j        dk    rt          d¦  «        ‚| j        \  }}}}|                      ||df¦  «        }t          j        |d¦  «                             ||df¦  «        }t          j	        |d¦  «                             ||df¦  «        }t          j
        |d¦  «                             t          j        ¦  «        }|dd…dd…d	f         |z  |dd…dd…d	f<   |dd…dd…df         |z  |dd…dd…df<   t          j        t          j
        |d¦  «        d
k    |d¦  «        }||fS )aì  Get keypoint predictions from score maps.

    Args:
        heatmaps (`np.ndarray` of shape `(batch_size, num_keypoints, height, width)`):
            Model predicted heatmaps.

    Returns:
        tuple: A tuple containing aggregated results.

        - coords (`np.ndarray` of shape `(batch_size, num_keypoints, 2)`):
            Predicted keypoint location.
        - scores (`np.ndarray` of shape `(batch_size, num_keypoints, 1)`):
            Scores (confidence) of the keypoints.
    zHeatmaps should be np.ndarrayr3   z Heatmaps should be 4-dimensionalr^   rH   r   )r   r   rH   Nr   g        )Ú
isinstancer8   ÚndarrayÚ	TypeErrorÚndimÚ
ValueErrorrh   ÚreshapeÚargmaxÚamaxÚtileÚastyper:   Úwhere)	rp   Ú
batch_sizeÚnum_keypointsÚ_r=   Úheatmaps_reshapedÚidxÚscoresÚpredss	            r*   Úget_keypoint_predictionsr…   ¯   sg  € õ �h¥¤
Ñ+Ô+ð 9ÝÐ7Ñ8Ô8Ð8Ø„}˜ÒÐÝÐ;Ñ<Ô<Ð<à*2¬.Ñ'€J�˜q %Ø ×(Ò(¨*°mÀRÐ)HÑIÔIÐÝ
Œ)Ð% qÑ
)Ô
)×
1Ò
1°:¸}ÈaÐ2PÑ
QÔ
Q€CÝŒWÐ&¨Ñ*Ô*×2Ò2°JÀÈqÐ3QÑRÔR€FåŒG�C˜Ñ#Ô#×*Ò*­2¬:Ñ6Ô6€EØ˜1˜1˜1˜a˜a˜a ˜7”^ eÑ+€Eˆ!ˆ!ˆ!ˆQˆQˆQ�ˆ'�NØ˜1˜1˜1˜a˜a˜a ˜7”^ uÑ,€Eˆ!ˆ!ˆ!ˆQˆQˆQ�ˆ'�NåŒH•R”W˜V YÑ/Ô/°#Ò5°u¸bÑAÔA€EØ�&ˆ=Ðr)   ÚcoordsÚbatch_heatmapsÚkernelc                 ó‚  ‡— |j         \  }}}}| j         d         }|dk    s||k    st          d¦  «        ‚t          |dz
  dz  ¦  «        Št          j        ˆfd„|D ¦   «         ¦  «        }t          j        |dd¦  «        }t          j        |¦  «        }t          j        |dd	¬
¦  «                             ¦   «         }| d         dz   | d         dz   |dz   z  z   }	|	|dz   |dz   z  t          j	        d||z  ¦  «         
                    d|¦  «        z  z  }	|	                     t          ¦  «         
                    dd¦  «        }	||	         }
||	dz            }||	|z   dz            }||	|z   dz            }||	|z
  dz
           }||	dz
           }||	dz
  |z
           }d||z
  z  }d||z
  z  }t          j        ||gd¬¦  «        }| 
                    ||dd¦  «        }|d|
z  z
  |z   }|d|
z  z
  |z   }d||z
  |z
  |
z   |
z   |z
  |z
  |z   z  }t          j        ||||gd¬¦  «        }| 
                    ||dd¦  «        }t          j                             |t          j        t          j        ¦  «        j        t          j        d¦  «        z  z   ¦  «        }| t          j        d||¦  «                             ¦   «         z  } | S )a  DARK post-processing. Implemented by unbiased_data_processing.

    Paper references:
    - Huang et al. The Devil is in the Details: Delving into Unbiased Data Processing for Human Pose Estimation (CVPR 2020).
    - Zhang et al. Distribution-Aware Coordinate Representation for Human Pose Estimation (CVPR 2020).

    Args:
        coords (`np.ndarray` of shape `(num_persons, num_keypoints, 2)`):
            Initial coordinates of human pose.
        batch_heatmaps (`np.ndarray` of shape `(batch_size, num_keypoints, height, width)`):
            Batched heatmaps as predicted by the model.
            A batch_size of 1 is used for the bottom up paradigm where all persons share the same heatmap.
            A batch_size of `num_persons` is used for the top down paradigm where each person has its own heatmaps.
        kernel (`int`, *optional*, defaults to 3):
            Gaussian kernel size (K) for modulation.

    Returns:
        `np.ndarray` of shape `(num_persons, num_keypoints, 2)` ):
            Refined coordinates.
    r   r   zQThe batch size of heatmaps should be 1 or equal to the batch size of coordinates.rH   c                 ó,   •— g | ]}ˆfd „|D ¦   «         ‘ŒS )c                 ó:   •— g | ]}t          |d ‰‰fd¬¦  «        ‘ŒS )gš™™™™™é?rJ   )ÚsigmaÚradiusÚaxes)r   )rZ   Úheatmapr�   s     €r*   r]   zApost_dark_unbiased_data_processing.<locals>.<listcomp>.<listcomp>ì   s0   ø€ ÐoÐoÐoÐ[b�_˜W¨C¸ÀÐ8HÈvÐVÑVÔVÐoÐoÐor)   r(   )rZ   rp   r�   s     €r*   r]   z6post_dark_unbiased_data_processing.<locals>.<listcomp>ë   s?   ø€ ð 	
ð 	
ð 	
àð pÐoÐoÐoÐfnÐoÑoÔoð	
ð 	
ð 	
r)   gü©ñÒMbP?é2   )rI   rI   rM   rM   Úedge)Úmode).r   ).r   r^   r   r4   re   zijmn,ijnk->ijmk)rh   rw   Úintr8   r9   ÚclipÚlogÚpadÚflattenÚarangerx   r|   ÚconcatenateÚlinalgr   Úfinfor:   ÚepsÚeyeÚeinsumÚsqueeze)r†   r‡   rˆ   r~   r   r>   r=   Ú
num_coordsÚbatch_heatmaps_padÚindexÚi_Úix1Úiy1Úix1y1Úix1_y1_Úix1_Úiy1_ÚdxÚdyÚ
derivativeÚdxxÚdyyÚdxyÚhessianr�   s                           @r*   Ú"post_dark_unbiased_data_processingr±   Ð   sÿ  ø€ ð* 0>Ô/CÑ,€J�˜v uØ”˜a”€JØ˜!ŠOˆO˜z¨ZÒ7Ð7ÝÐlÑmÔmÐmÝ�&˜1‘* Ñ"Ñ#Ô#€FÝ”Xð	
ð 	
ð 	
ð 	
à*ð	
ñ 	
ô 	
ñô €Nõ ”W˜^¨U°BÑ7Ô7€NÝ”V˜NÑ+Ô+€Nåœ Ð0PÐW]Ð^Ñ^Ô^×fÒfÑhÔhÐØ�6ŒN˜QÑ &¨¤.°1Ñ"4¸À¹Ñ!CÑC€EØ	ˆe�a‰i˜F Q™JÑ'­"¬)°A°zÀMÑ7QÑ*RÔ*R×*ZÒ*ZÐ[]Ð_lÑ*mÔ*mÑmÑm€EØ�LŠL�ÑÔ×%Ò% b¨!Ñ,Ô,€EØ	˜EÔ	"€BØ
˜U Q™YÔ
'€CØ
˜U U™]¨QÑ.Ô
/€CØ˜u u™}¨qÑ0Ô1€EØ  ¨¡°Ñ!2Ô3€GØ˜e a™iÔ(€DØ˜e a™i¨%Ñ/Ô0€DØ	��d‘
Ñ	€BØ	��d‘
Ñ	€BÝ”  R ¨qÐ1Ñ1Ô1€JØ×#Ò# J°¸qÀ!ÑDÔD€JØ
��B‘‰,˜Ñ
€CØ
��B‘‰,˜Ñ
€CØ
�˜‘˜sÑ" RÑ'¨"Ñ,¨tÑ3°dÑ:¸WÑDÑ
E€CÝŒn˜c 3¨¨SÐ1¸Ð:Ñ:Ô:€GØ�oŠo˜j¨-¸¸AÑ>Ô>€GÝŒi�mŠm˜G¥b¤h­r¬zÑ&:Ô&:Ô&>ÅÄÈÁÄÑ&JÑJÑKÔK€GØ
�bŒiÐ)¨7°JÑ?Ô?×GÒGÑIÔIÑI€FØ€Mr)   r@   rA   Úoutput_sizec                 ó<  — | j         d         dvrt          d¦  «        ‚t          |¦  «        dk    rt          d¦  «        ‚t          |¦  «        dk    rt          d¦  «        ‚t          |¦  «        dk    rt          d¦  «        ‚|dz  }|d         |d	         d
z
  z  }|d	         |d         d
z
  z  }t          j        | ¦  «        }| dd…d	f         |z  |d	         z   |d	         dz  z
  |dd…d	f<   | dd…df         |z  |d         z   |d         dz  z
  |dd…df<   |S )ao  Get final keypoint predictions from heatmaps and apply scaling and
    translation to map them back to the image.

    Note:
        num_keypoints: K

    Args:
        coords (`np.ndarray` of shape `(num_keypoints, ndims)`):

            * If ndims=2, corrds are predicted keypoint location.
            * If ndims=4, corrds are composed of (x, y, scores, tags)
            * If ndims=5, corrds are composed of (x, y, scores, tags,
              flipped_tags)

        center (`np.ndarray` of shape `(2,)`):
            Center of the bounding box (x, y).
        scale (`np.ndarray` of shape `(2,)`):
            Scale of the bounding box wrt original image of width and height.
        output_size (`np.ndarray` of shape `(2,)`):
            Size of the destination heatmaps in (height, width) format.

    Returns:
        np.ndarray: Predicted coordinates in the images.
    r   )rH   r3   é   z5Coordinates need to have either 2, 4 or 5 dimensions.rH   z9Center needs to have 2 elements, one for x and one for y.z,Scale needs to consist of a width and heightz2Output size needs to consist of a height and widthr,   r   r7   Nr4   )rh   rw   Úlenr8   Ú	ones_like)r†   r@   rA   r²   rV   rU   Útarget_coordss          r*   Útransform_predsr¸     sF  € ð2 „|�A„˜iÐ'Ð'ÝÐPÑQÔQÐQÝ
ˆ6�{„{�aÒÐÝÐTÑUÔUÐUÝ
ˆ5�z„z�Q‚€ÝÐGÑHÔHÐHÝ
ˆ;ÑÔ˜1ÒÐÝÐMÑNÔNÐNð �E‰M€Eð �AŒh˜+ aœ.¨3Ñ.Ñ/€GØ�AŒh˜+ aœ.¨3Ñ.Ñ/€Gå”L Ñ(Ô(€MØ     A œ,¨Ñ0°6¸!´9Ñ<¸uÀQ¼xÈ#¹~ÑM€M�!�!�!�Q�$ÑØ     A œ,¨Ñ0°6¸!´9Ñ<¸uÀQ¼xÈ#¹~ÑM€M�!�!�!�Q�$ÑàÐr)   Úbboxesc                 ó’   — | dd…df         | dd…df         z   dz
  | dd…df<   | dd…df         | dd…df         z   dz
  | dd…df<   | S )a¥  
    Converts bounding boxes from the COCO format to the Pascal VOC format.

    In other words, converts from (top_left_x, top_left_y, width, height) format
    to (top_left_x, top_left_y, bottom_right_x, bottom_right_y).

    Args:
        bboxes (`np.ndarray` of shape `(batch_size, 4)):
            Bounding boxes in COCO format.

    Returns:
        `np.ndarray` of shape `(batch_size, 4) in Pascal VOC format.
    NrH   r   r   r   r(   )r¹   s    r*   Úcoco_to_pascal_vocr»   <  so   € ð ˜!˜!˜!˜Q˜$”< &¨¨¨¨A¨¤,Ñ.°Ñ2€Fˆ1ˆ1ˆ1ˆaˆ4�LØ˜!˜!˜!˜Q˜$”< &¨¨¨¨A¨¤,Ñ.°Ñ2€Fˆ1ˆ1ˆ1ˆaˆ4�Là€Mr)   c            '       ó  ‡ — e Zd ZdZeZdgZeZe	Z
dddœZdZdZdZdZdee         fˆ fd	„Zed
edeeee                           ej        z  dee         defˆ fd„¦   «         Z	 d6d
edeeee                           ej        z  dz  dededeedf         dz  defd„Zej         j!        ddde"e         de"e         dede#ddfd„¦   «         Z$	 	 	 d7d
ed         dede#dddede#d ed!ed"ed#eee         z  dz  d$eee         z  dz  d%edz  d&e#dz  d'edz  d(ee%z  dz  d)ed*edeej        z  dz  def&d+„Z&	 d8d-ej        dej        dej        d.e'fd/„Z(	 	 	 d9d0d1deeee                           ej        z  d2e'd3edz  d4e%ee"         z  dz  f
d5„Z)ˆ xZ*S ):ÚVitPoseImageProcessorz6Torchvision backend for VitPose with affine transform.Úpixel_valuesé   éÀ   )r>   r=   Tr,   Úkwargsc                 ó:   •—  t          ¦   «         j        di |¤Ž d S )Nr(   )ÚsuperÚ__init__)ÚselfrÁ   Ú	__class__s     €r*   rÄ   zVitPoseImageProcessor.__init___  s&   ø€ Ø�‰ŒÔÐ"Ð"˜6Ð"Ð"Ð"Ð"Ð"r)   ÚimagesÚboxesrq   c                 ó:   •—  t          ¦   «         j        ||fi |¤ŽS )a  
        boxes (`list[list[list[float]]]` or `np.ndarray`):
            List or array of bounding boxes for each image. Each box should be a list of 4 floats representing the
            bounding box coordinates in COCO format (top_left_x, top_left_y, width, height).
        )rÃ   Ú
preprocess)rÅ   rÇ   rÈ   rÁ   rÆ   s       €r*   rÊ   z VitPoseImageProcessor.preprocessb  s&   ø€ ð "�u‰wŒwÔ! &¨%Ð:Ð:°6Ð:Ð:Ð:r)   NÚdo_convert_rgbÚinput_data_formatÚdeviceztorch.devicec                 óZ   — |                       ||||¬¦  «        }||d<    | j        |fi |¤ŽS )z"Handle extra inputs beyond images.)rÇ   rË   rÌ   rÍ   rÈ   )Ú_prepare_image_like_inputsÚ_preprocess)rÅ   rÇ   rÈ   rË   rÌ   rÍ   rÁ   s          r*   Ú_preprocess_image_like_inputsz3VitPoseImageProcessor._preprocess_image_like_inputsp  sO   € ð ×0Ò0Ø¨.ÐL]Ðflð 1ñ 
ô 
ˆð  ˆˆw‰ØˆtÔ Ð1Ð1¨&Ð1Ð1Ð1r)   Úimageztorch.Tensorr@   rA   Úrotationrd   c                 ó¬  — t          ||dz  t          j        |j        |j        f¦  «        dz
  |dz  ¦  «        }|                     ddd¦  «                             ¦   «                              ¦   «         }t          |||j        |j        f¬¦  «        }t          j
        |¦  «                             ddd¦  «                             |j        ¦  «        }	|	S )z7Apply an affine transformation to a torch tensor image.g       @r7   r,   r   rH   r   )r\   rk   rd   )rW   r8   r9   r=   r>   ÚpermuteÚcpuÚnumpyro   ÚtorchÚ
from_numpyÚtorÍ   )
rÅ   rÒ   r@   rA   rÓ   rd   ÚtransformationÚimage_npÚtransformed_npÚtransformeds
             r*   r   z&VitPoseImageProcessor.affine_transform�  sÅ   € õ )Ø�f˜s‘l¥B¤H¨d¬j¸$¼+Ð-FÑ$GÔ$GÈ#Ñ$MÈuÐW\É}ñ
ô 
ˆð —=’=  A qÑ)Ô)×-Ò-Ñ/Ô/×5Ò5Ñ7Ô7ˆÝ*¨x¸>ÐQUÔQ\Ð^bÔ^hÐPiÐjÑjÔjˆÝÔ& ~Ñ6Ô6×>Ò>¸qÀ!ÀQÑGÔG×JÒJÈ5Ì<ÑXÔXˆØÐr)   Ú	do_resizeÚresamplez7PILImageResampling | tvF.InterpolationMode | int | NoneÚdo_center_cropÚ	crop_sizeÚ
do_rescaleÚrescale_factorÚdo_normalizeÚ
image_meanÚ	image_stdÚdo_padÚpad_sizeÚdisable_groupingÚreturn_tensorsr   r    c           	      óÊ  — |�p|rng }t          ||¦  «        D ]Y\  }}|D ]Q}t          ||j        |j        |¬¦  «        \  }}|                      |||d|¬¦  «        }|                     |¦  «         ŒRŒZ|}t          ||¬¦  «        \  }}i }|                     ¦   «         D ]$\  }}|                      ||||	|
|¦  «        }|||<   Œ%t          ||¦  «        } t          d| i|¬¦  «        S )z!Custom preprocessing for VitPose.N)r/   r0   r    r   )rÓ   rd   )rê   r¾   )ÚdataÚtensor_type)ÚziprB   r=   r>   r   Úappendr	   ÚitemsÚrescale_and_normalizer
   r   )!rÅ   rÇ   rß   rd   rà   rá   râ   rã   rä   rå   ræ   rç   rè   ré   rê   rë   r   r    rÈ   rÁ   Útransformed_imagesrÒ   Úimage_boxesr.   r@   rA   Útransformed_imageÚgrouped_imagesÚgrouped_images_indexÚprocessed_images_groupedrh   Ústacked_imagesÚprocessed_imagess!                                    r*   rÐ   z!VitPoseImageProcessor._preprocess“  sU  € ð. ÐÐ!4ÐØ!#ÐÝ&)¨&°%Ñ&8Ô&8ð 	Að 	AÑ"��{Ø&ð Að A�CÝ$;ØØ$(¤JØ%)¤[Ø)9ð	%ñ %ô %‘M�F˜Eð )-×(=Ò(=¸eÀVÈUÐ]^ÐeiÐ(=Ñ(jÔ(jÐ%Ø&×-Ò-Ð.?Ñ@Ô@Ð@Ð@ðAð (ˆFå/DÀVÐ^nÐ/oÑ/oÔ/oÑ,ˆÐ,Ø#%Ð Ø%3×%9Ò%9Ñ%;Ô%;ð 	=ð 	=Ñ!ˆE�>Ø!×7Ò7Ø 
¨N¸LÈ*ÐV_ñô ˆNð /=Ð$ UÑ+Ð+Ý)Ð*BÐDXÑYÔYÐÝ .Ð2BÐ!CÐQ_Ð`Ñ`Ô`Ð`r)   é   rp   rˆ   c                 óâ   — |j         \  }}}}t          |¦  «        \  }	}
t          |	||¬¦  «        }t          |¦  «        D ],}t	          ||         ||         ||         ||g¬¦  «        ||<   Œ-||
fS )zRGet final keypoint predictions from heatmaps and transform them back to the image.©rˆ   )r@   rA   r²   )rh   r…   r±   rg   r¸   )rÅ   rp   r@   rA   rˆ   r~   r€   r>   r=   r†   rƒ   r„   r[   s                r*   Úkeypoints_from_heatmapsz-VitPoseImageProcessor.keypoints_from_heatmapsÂ  s‹   € ð (0¤~Ñ$ˆ
�A�v˜uÝ1°(Ñ;Ô;‰ˆ�Ý2°6¸8ÈFÐSÑSÔSˆÝ�zÑ"Ô"ð 	pð 	pˆAÝ& u¨Q¤x¸¸q¼	ÈÈqÌÐ`fÐhmÐ_nÐoÑoÔoˆE�!‰HˆHØ�fˆ}Ðr)   Úoutputsr   Úkernel_sizeÚ	thresholdÚtarget_sizesc                 ó   — t          | d¦  «         |j        j        \  }}}}|�"|t          |¦  «        k    rt	          d¦  «        ‚t          j        |dft
          j        ¬¦  «        }	t          j        |dft
          j        ¬¦  «        }
t          t          j
        |Ž ¦  «        }t          |¦  «        D ]�}|�B||         d         ||         d         }}t          j        ||||g¦  «        }||         |z  ||<   | j        d         | j        d	         }}t          ||         ||¬
¦  «        \  }}||	|dd…f<   ||
|dd…f<   ŒŽ|                      |j                             ¦   «                              ¦   «         |	|
|¬¦  «        \  }}t          j        |dft
          j        ¬¦  «        }|	dd…dd…f         |dd…dd…f<   |
dd…dd…f         |dd…dd…f<   t%          j        |¦  «        }t%          j        |¦  «        }t%          j        d|¦  «        }t%          j        t+          |¦  «        ¦  «        }g }t-          |||¦  «        }|D ]ƒ}g }|D ]g}t/          |¦  «        \  }}} |                     ¦   «         }|}!|�||k    }"||"         }||"         }|!|"         }!|||!| dœ}#|                     |#¦  «         Œh|                     |¦  «         Œ„|S )aî  
        Transform the heatmaps into keypoint predictions and transform them back to the image.

        Args:
            outputs (`VitPoseEstimatorOutput`):
                VitPoseForPoseEstimation model outputs.
            boxes (`list[list[list[float]]]` or `np.ndarray`):
                List or array of bounding boxes for each image. Each box should be a list of 4 floats representing the bounding
                box coordinates in COCO format (top_left_x, top_left_y, width, height).
            kernel_size (`int`, *optional*, defaults to 11):
                Gaussian kernel size (K) for modulation.
            threshold (`float`, *optional*, defaults to None):
                Score threshold to keep object detection predictions.
            target_sizes (`torch.Tensor` 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 unset, predictions will be resize with the default value.
        Returns:
            `list[list[Dict]]`: A list of dictionaries, each dictionary containing the keypoints and boxes for an image
            in the batch as predicted by the model.
        rØ   NzTMake sure that you pass in as many target sizes as the batch dimension of the logitsrH   r5   r   r   r=   r>   )r/   r0   rý   r3   )Ú	keypointsrƒ   ÚlabelsÚbbox)r   rp   rh   rµ   rw   r8   rP   r:   ÚlistÚ	itertoolsÚchainrg   r9   rd   rB   rþ   rÖ   r×   rØ   Útensorr˜   r»   rï   ÚnextrŸ   rð   )$rÅ   rÿ   rÈ   r   r  r  r~   r   r€   ÚcentersÚscalesÚflattened_boxesr[   r/   r0   Úscale_factorr=   r>   r@   rA   r„   rƒ   Ú	all_boxesÚposesr  Úbboxes_xyxyÚresultsÚpose_bbox_pairsÚimage_bboxesÚimage_resultsÚposeÚscoreÚ	bbox_xyxyÚkeypoints_labelsÚkeepÚpose_results$                                       r*   Úpost_process_pose_estimationz2VitPoseImageProcessor.post_process_pose_estimationÑ  s  € õ8 	˜$ Ñ(Ô(Ð(Ø*1Ô*:Ô*@Ñ'ˆ
�M 1 aØÐ#¨
µc¸,Ñ6GÔ6GÒ(GÐ(GÝÐsÑtÔtÐtÝ”(˜J¨˜?µ"´*Ð=Ñ=Ô=ˆÝ”˜: q˜/µ´Ð<Ñ<Ô<ˆÝ�yœ°Ð6Ñ7Ô7ˆÝ�zÑ"Ô"ð 	!ð 	!ˆAØÐ'Ø,8¸¬O¸AÔ,>ÀÈQÄÐPQÔ@R˜\�Ý!œx¨°lÀKÐQ]Ð(^Ñ_Ô_�Ø%4°QÔ%7¸,Ñ%F� Ñ"Ø œI gÔ.°´	¸(Ô0C�6ˆEÝ3°OÀAÔ4FÐTYÐhnÐoÑoÔo‰MˆF�EØ"ˆG�A�q�q�q�D‰MØ ˆF�1�a�a�a�4‰LˆLØ×4Ò4ØÔ× Ò Ñ"Ô"×(Ò(Ñ*Ô*¨G°VÀKð 5ñ 
ô 
‰ˆˆvõ ”H˜j¨!˜_µB´JÐ?Ñ?Ô?ˆ	Ø# A A A q¨ s FœOˆ	�!�!�!�Q�q�S�&ÑØ" 1 1 1 a¨ c 6œNˆ	�!�!�!�Q�q�S�&ÑÝ”˜UÑ#Ô#ˆÝ”˜fÑ%Ô%ˆÝ”˜a Ñ/Ô/ˆÝ”lÕ#5°iÑ#@Ô#@ÑAÔAˆØ79ˆÝ˜e V¨[Ñ9Ô9ˆØ!ð 	*ð 	*ˆLØ;=ˆMØ!ð 
2ð 
2�Ý)-¨oÑ)>Ô)>Ñ&��e˜YØŸš™œ�Ø#)Ð ØÐ(Ø  9Ò,�DØ œ:�DØ! $œK�EØ'7¸Ô'=Ð$Ø,0¸EÐM]ÐgpÐqÐq�Ø×$Ò$ [Ñ1Ô1Ð1Ð1Ø�NŠN˜=Ñ)Ô)Ð)Ð)Øˆr)   )N)Tr,   N)rû   )rû   NN)+r!   r"   r#   r$   r   Úvalid_kwargsÚmodel_input_namesr   ræ   r   rç   rd   rã   rå   r   r    r   rÄ   r   r   r  r'   r8   rt   r   rÊ   r%   r   r   ÚstrrÑ   rØ   ÚcompilerÚdisableÚtupler   r   r   rÐ   r“   rþ   r  Ú__classcell__)rÆ   s   @r*   r½   r½   P  sð  ø€ € € € € à@Ð@à.€LØ'Ð(Ðà&€JØ$€IØ CÐ(Ð(€DØ€JØ€LØÐØÐð# Ð(CÔ!Dð #ð #ð #ð #ð #ð #ð ð;àð;ð �D˜˜eœÔ%Ô&¨¬Ñ3ð;ð Ð4Ô5ð	;ð
 
ð;ð ;ð ;ð ;ð ;ñ „^ð;ð& 59ð2ð 2àð2ð �D˜˜eœÔ%Ô&¨¬Ñ3°dÑ:ð2ð ð	2ð
 ,ð2ð �c˜>Ð)Ô*¨TÑ1ð2ð 
ð2ð 2ð 2ð 2ð" „^Ôðàðð �e”ðð �UŒ|ð	ð
 ðð ðð 
ðð ð ñ ÔððD %)Ø"'Ø*.ð'-að -aà�^Ô$ð-að ð-að ð	-að
 Lð-að ð-að ð-að ð-að ð-að ð-að ˜D œKÑ'¨$Ñ.ð-að ˜4 œ;Ñ&¨Ñ-ð-að �t‘ð-að ˜T‘/ð-að  ™+ð-að  ˜jÑ(¨4Ñ/ð!-að" "ð#-að$  ð%-að& �b”jÑ  4Ñ'ð'-að* 
ð+-að -að -að -aðh ðð à”*ðð ”
ðð Œzð	ð
 ðð ð ð ð& Ø"&Ø8<ðFð Fà)ðFð �D˜˜eœÔ%Ô&¨¬Ñ3ðFð ð	Fð
 ˜4‘<ðFð ! 4¨¤;Ñ.°Ñ5ðFð Fð Fð Fð Fð Fð Fð Fr)   r½   )r,   r-   )r   )<r$   r  rQ   Útypingr   r   r×   r8   rØ   Útorchvision.transforms.v2r   ÚtvFÚimage_processing_backendsr   Úimage_processing_utilsr   Úimage_transformsr	   r
   Úimage_utilsr   r   r   r   r   r   Úprocessing_utilsr   r   Úutilsr   r   r   r   r   Úscipy.linalgr   Úscipy.ndimager   r   Úmodeling_vitposer   Ú
get_loggerr!   Úloggerr   r#  r  rt   r“   r'   rB   rW   ro   r…   r±   r¸   r»   r½   Ú__all__r(   r)   r*   ú<module>r4     s·  ðð )Ð (à Ð Ð Ð Ø €€€Ø 'Ð 'Ð 'Ð 'Ð 'Ð 'Ð 'Ð 'à Ð Ð Ð Ø €€€Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7à ;Ð ;Ð ;Ð ;Ð ;Ð ;Ø 2Ð 2Ð 2Ð 2Ð 2Ð 2Ø EÐ EÐ EÐ EÐ EÐ EÐ EÐ Eðð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð 5Ð 4Ð 4Ð 4Ð 4Ð 4Ð 4Ð 4ðð ð ð ð ð ð ð ð ð ð ð ð ð ð ÐÑÔð @Ø Ð Ð Ð Ð Ð Ø?Ð?Ð?Ð?Ð?Ð?Ð?Ð?àð 9Ø8Ð8Ð8Ð8Ð8Ð8à	ˆÔ	˜HÑ	%Ô	%€ð	#ð 	#ð 	#ð 	#ð 	# ,°eð 	#ñ 	#ô 	#ð 	#ð" $Ø ð)ð )Ø	�‰˜œ
Ñ	"ð)àð)ð ð)ð ð	)ð
 ð)ð )ð )ð )ðX"˜5ð "¨b¬jð "ÀBÄJð "Ð]_Ô]gð "ð "ð "ð "ðJð ð ð4 r¤zð °e¸B¼JÈÌ
Ð<RÔ6Sð ð ð ð ðB9ð 9¨r¬zð 9È2Ì:ð 9Ð_bð 9ÐkmÔkuð 9ð 9ð 9ð 9ðx-˜BœJð -°´
ð -À2Ä:ð -Ð\^Ô\fð -ÐkmÔkuð -ð -ð -ð -ð`˜rœzð ¨b¬jð ð ð ð ð( ðFð Fð Fð Fð FÐ.ñ Fô Fñ „ðFðR #Ð
#€€€r)   