§
    ­¸j1,  ã                  ó²   — d Z ddlmZ ddlZddlmZ ddlmZ ddlZddlm	Z	m
Z
 ddlmZ dZ G d	„ d
e
j        ¦  «        Z G d„ de
j        ¦  «        Zdd„ZdS )z1Various positional encodings for the transformer.é    )ÚannotationsN)ÚCallable)Úcast)ÚTensorÚnn)ÚNestedTensoré   c                  óp   ‡ — e Zd ZdZ	 	 	 	 d!d"ˆ fd„Zd#d$ˆ fd„Z	 d#d%ˆ fd„Zd&d„Zd#d'd„Zd'd„Z	d#d(d „Z
ˆ xZS ))ÚPositionEmbeddingSinez¢This is a more standard version of the position embedding, very similar to the one used by the Attention is all
    you need paper, generalized to work on images.é@   é'  FNÚnum_pos_featsÚintÚtemperatureÚ	normalizeÚboolÚscaleúfloat | NoneÚreturnÚNonec                óè   •— t          ¦   «                              ¦   «          || _        || _        || _        |�|du rt          d¦  «        ‚|€dt          j        z  }|| _        d| _	        i | _
        d S )NFz+normalize should be True if scale is passedé   )ÚsuperÚ__init__r   r   r   Ú
ValueErrorÚmathÚpir   Ú_exportÚ
_pos_cache)Úselfr   r   r   r   Ú	__class__s        €úo/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/rfdetr/models/position_encoding.pyr   zPositionEmbeddingSine.__init__%   s~   ø€ õ 	‰Œ×ÒÑÔÐØ*ˆÔØ&ˆÔØ"ˆŒØÐ ¨eÐ!3Ð!3ÝÐJÑKÔKÐKØˆ=Ø�œ‘KˆEØˆŒ
ØˆŒð lnˆŒˆˆó    TÚmodeú'PositionEmbeddingSine'c                ó|   •— |r| j                              ¦   «          t          ¦   «                              |¦  «        S )zäSet training mode and release evaluation-only cached tensors before training.

        Args:
            mode: Whether to enable training mode.

        Returns:
            This module after updating its training mode.
        )r   Úclearr   Útrain)r    r$   r!   s     €r"   r(   zPositionEmbeddingSine.train:   s7   ø€ ð ð 	$ØŒO×!Ò!Ñ#Ô#Ð#Ý‰wŒw�}Š}˜TÑ"Ô"Ð"r#   ÚfnúCallable[[Tensor], Tensor]Úrecursec                óò   •— | j                              ¦   «          t          t          t          t          gt          f         t
          gt          f         t          ¦   «         j        ¦  «        } |||¦  «        S )aF  Clear cached tensors before applying a device or dtype transformation.

        Args:
            fn: Transformation applied by :class:`torch.nn.Module`.
            recurse: Whether to apply the transformation recursively to child modules.

        Returns:
            This module after applying the transformation.
        )	r   r'   r   r   r   r   r   r   Ú_apply)r    r)   r+   Úapplyr!   s       €r"   r-   zPositionEmbeddingSine._applyG   sd   ø€ ð 	Œ×ÒÑÔÐÝÝ•h¥˜x­Ð/Ô0µ$Ð7Õ9NÐNÔOÝ‰GŒGŒNñ
ô 
ˆð ˆu�R˜Ñ!Ô!Ð!r#   c                óD   — d| _         | j        | _        | j        | _        d S )NT)r   ÚforwardÚ_forward_originÚforward_export©r    s    r"   ÚexportzPositionEmbeddingSine.export\   s!   € ØˆŒØ#œ|ˆÔØÔ*ˆŒˆˆr#   Útensor_listr   Úalign_dim_ordersr   c                óÌ  — |j         }|€J ‚|j        r½| j        s¶t          |j        ¦  «        |j        || j        | j        | j        | j	        f}| j
                             |¦  «        }|€d|                      ||¦  «        }t          | j
        ¦  «        t          k    r'| j
        t          t!          | j
        ¦  «        ¦  «        = || j
        |<   |S |                      ||¦  «        S ©N)ÚmaskÚ
no_paddingÚtrainingÚtupleÚshapeÚdevicer   r   r   r   r   ÚgetÚ_computeÚlenÚ_POS_CACHE_MAX_ENTRIESÚnextÚiter)r    r5   r6   r9   ÚkeyÚcacheds         r"   r0   zPositionEmbeddingSine.forwarda   sæ   € ØÔˆØÐÐÐØÔ!ð 	¨$¬-ð 	õ �d”jÑ!Ô!Ø”Ø ØÔ"ØÔ Ø”Ø”
ðˆCð ”_×(Ò(¨Ñ-Ô-ˆFØˆ~ØŸš {Ð4DÑEÔE�Ý�t”Ñ'Ô'Õ+AÒAÐAàœ­­T°$´/Ñ-BÔ-BÑ(CÔ(CÐDØ'-�” Ñ$ØˆMØ�}Š}˜[Ð*:Ñ;Ô;Ð;r#   c           
     ó®  — |j         }|j        }|€J ‚| }|                     dt          j        ¬¦  «        }|                     dt          j        ¬¦  «        }| j        r@d}||dd…dd…dd…f         |z   z  | j        z  }||dd…dd…dd…f         |z   z  | j        z  }t          j        | j        t          j        |j	        ¬¦  «        }	| j
        d|	dz  z  | j        z  z  }	|dd…dd…dd…df         |	z  }
|dd…dd…dd…df         |	z  }t          j        |
dd…dd…dd…ddd…f                              ¦   «         |
dd…dd…dd…ddd…f                              ¦   «         fd	¬
¦  «                             d¦  «        }
t          j        |dd…dd…dd…ddd…f                              ¦   «         |dd…dd…dd…ddd…f                              ¦   «         fd	¬
¦  «                             d¦  «        }|r/t          j        ||
fd¬
¦  «                             dddd¦  «        }n.t          j        ||
fd¬
¦  «                             dddd¦  «        }|S )a©  Compute a sine/cosine position embedding without consulting the cache.

        Args:
            tensor_list: Batch tensor and padding mask used to derive spatial coordinates.
            align_dim_orders: Whether to return spatial dimensions before batch and channels.

        Returns:
            Position embedding shaped ``(H, W, B, C)`` when *align_dim_orders* is true,
            otherwise ``(B, C, H, W)``.
        Né   ©Údtyper   ç�íµ ÷Æ°>éÿÿÿÿ©rJ   r>   r   é   ©Údimé   )Útensorsr9   ÚcumsumÚtorchÚfloat32r   r   Úaranger   r>   r   ÚstackÚsinÚcosÚflattenÚcatÚpermute)r    r5   r6   Úxr9   Únot_maskÚy_embedÚx_embedÚepsÚdim_tÚpos_xÚpos_yÚposs                r"   r@   zPositionEmbeddingSine._compute   s¨  € ð ÔˆØÔˆØÐÐÐØ�5ˆØ—/’/ !­5¬=�/Ñ9Ô9ˆØ—/’/ !­5¬=�/Ñ9Ô9ˆØŒ>ð 	HØˆCØ ¨¨¨¨B¨C¨C°°°¨Ô!3°cÑ!9Ñ:¸T¼ZÑGˆGØ ¨¨¨¨A¨A¨A¨r¨s¨s¨Ô!3°cÑ!9Ñ:¸T¼ZÑGˆGå”˜TÔ/µu´}ÈQÌXÐVÑVÔVˆØÔ  Q¨%°1©*Ñ%5¸Ô8JÑ%JÑKˆà˜˜˜˜1˜1˜1˜a˜a˜a ˜Ô&¨Ñ.ˆØ˜˜˜˜1˜1˜1˜a˜a˜a ˜Ô&¨Ñ.ˆÝ”˜U 1 1 1 a a a¨¨¨¨A¨D¨q¨D =Ô1×5Ò5Ñ7Ô7¸¸q¸q¸qÀ!À!À!ÀQÀQÀQÈÈÈ1È¸}Ô9M×9QÒ9QÑ9SÔ9SÐTÐZ[Ð\Ñ\Ô\×dÒdÐefÑgÔgˆÝ”˜U 1 1 1 a a a¨¨¨¨A¨D¨q¨D =Ô1×5Ò5Ñ7Ô7¸¸q¸q¸qÀ!À!À!ÀQÀQÀQÈÈÈ1È¸}Ô9M×9QÒ9QÑ9SÔ9SÐTÐZ[Ð\Ñ\Ô\×dÒdÐefÑgÔgˆØð 	GÝ”)˜U E˜N°Ð2Ñ2Ô2×:Ò:¸1¸aÀÀAÑFÔFˆCˆCõ ”)˜U E˜N°Ð2Ñ2Ô2×:Ò:¸1¸aÀÀAÑFÔFˆCàˆ
r#   r9   c           
     ó’  — |€J ‚| }|                      dt          j        ¬¦  «        }|                      dt          j        ¬¦  «        }| j        r@d}||d d …dd …d d …f         |z   z  | j        z  }||d d …d d …dd …f         |z   z  | j        z  }t          j        | j        t          j        |j        ¬¦  «        }| j        d|dz  z  | j        z  z  }|d d …d d …d d …d f         |z  }|d d …d d …d d …d f         |z  }	t          j	        |d d …d d …d d …dd d…f          
                    ¦   «         |d d …d d …d d …dd d…f                              ¦   «         fd¬	¦  «                             d
¦  «        }t          j	        |	d d …d d …d d …dd d…f          
                    ¦   «         |	d d …d d …d d …dd d…f                              ¦   «         fd¬	¦  «                             d
¦  «        }	|r/t          j        |	|fd
¬	¦  «                             dddd
¦  «        }
n.t          j        |	|fd
¬	¦  «                             dd
dd¦  «        }
|
S )NrH   rI   r   rK   rL   rM   r   rN   rO   rQ   )rS   rT   rU   r   r   rV   r   r>   r   rW   rX   rY   rZ   r[   r\   )r    r9   r6   r^   r_   r`   ra   rb   rc   rd   re   s              r"   r2   z$PositionEmbeddingSine.forward_export¤   s–  € ØÐÐÐØ�5ˆØ—/’/ !­5¬=�/Ñ9Ô9ˆØ—/’/ !­5¬=�/Ñ9Ô9ˆØŒ>ð 	HØˆCØ ¨¨¨¨B¨C¨C°°°¨Ô!3°cÑ!9Ñ:¸T¼ZÑGˆGØ ¨¨¨¨A¨A¨A¨r¨s¨s¨Ô!3°cÑ!9Ñ:¸T¼ZÑGˆGå”˜TÔ/µu´}ÈTÌ[ÐYÑYÔYˆØÔ  Q¨%°1©*Ñ%5¸Ô8JÑ%JÑKˆà˜˜˜˜1˜1˜1˜a˜a˜a ˜Ô&¨Ñ.ˆØ˜˜˜˜1˜1˜1˜a˜a˜a ˜Ô&¨Ñ.ˆÝ”˜U 1 1 1 a a a¨¨¨¨A¨D¨q¨D =Ô1×5Ò5Ñ7Ô7¸¸q¸q¸qÀ!À!À!ÀQÀQÀQÈÈÈ1È¸}Ô9M×9QÒ9QÑ9SÔ9SÐTÐZ[Ð\Ñ\Ô\×dÒdÐefÑgÔgˆÝ”˜U 1 1 1 a a a¨¨¨¨A¨D¨q¨D =Ô1×5Ò5Ñ7Ô7¸¸q¸q¸qÀ!À!À!ÀQÀQÀQÈÈÈ1È¸}Ô9M×9QÒ9QÑ9SÔ9SÐTÐZ[Ð\Ñ\Ô\×dÒdÐefÑgÔgˆØð 	GÝ”)˜U E˜N°Ð2Ñ2Ô2×:Ò:¸1¸aÀÀAÑFÔFˆCˆCõ ”)˜U E˜N°Ð2Ñ2Ô2×:Ò:¸1¸aÀÀAÑFÔFˆCàˆ
r#   )r   r   FN)
r   r   r   r   r   r   r   r   r   r   )T)r$   r   r   r%   )r)   r*   r+   r   r   r%   ©r   r   )r5   r   r6   r   r   r   )r9   r   r6   r   r   r   )Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   r(   r-   r4   r0   r@   r2   Ú__classcell__©r!   s   @r"   r   r   !   s  ø€ € € € € ð6ð 6ð
  Ø ØØ"ðnð nð nð nð nð nð nð*#ð #ð #ð #ð #ð #ð #ð  ð"ð "ð "ð "ð "ð "ð "ð*+ð +ð +ð +ð
<ð <ð <ð <ð <ð<#ð #ð #ð #ðJð ð ð ð ð ð ð ð r#   r   c                  ó>   ‡ — e Zd ZdZddˆ fd„Zdd„Zdd	„Zdd„Zˆ xZS )ÚPositionEmbeddingLearnedz Absolute pos embedding, learned.é   r   r   r   r   c                óæ   •— t          ¦   «                              ¦   «          t          j        d|¦  «        | _        t          j        d|¦  «        | _        |                      ¦   «          d| _        d S )Né2   F)r   r   r   Ú	EmbeddingÚ	row_embedÚ	col_embedÚreset_parametersr   )r    r   r!   s     €r"   r   z!PositionEmbeddingLearned.__init__Á   s\   ø€ Ý‰Œ×ÒÑÔÐÝœ b¨-Ñ8Ô8ˆŒÝœ b¨-Ñ8Ô8ˆŒØ×ÒÑÔÐØˆŒˆˆr#   c                ó   — t           ‚r8   )ÚNotImplementedErrorr3   s    r"   r4   zPositionEmbeddingLearned.exportÈ   s   € Ý!Ð!r#   c                óª   — t           j                             | j        j        ¦  «         t           j                             | j        j        ¦  «         d S r8   )r   ÚinitÚuniform_rt   Úweightru   r3   s    r"   rv   z)PositionEmbeddingLearned.reset_parametersË   s>   € Ý
Œ×Ò˜œÔ.Ñ/Ô/Ð/Ý
Œ×Ò˜œÔ.Ñ/Ô/Ð/Ð/Ð/r#   r5   r   r   c                ó0  — |j         }|j        d d…         \  }}t          j        ||j        ¬¦  «        }t          j        ||j        ¬¦  «        }|                      |¦  «        }|                      |¦  «        }t          j        |                     d¦  «         	                    |dd¦  «        |                     d¦  «         	                    d|d¦  «        gd¬¦  «                             d¦  «         	                    dd|j        d         d¦  «        }	|	S )Nr   )r>   r   rH   rL   rO   )
rR   r=   rT   rV   r>   ru   rt   r[   Ú	unsqueezeÚrepeat)
r    r5   r]   ÚhÚwÚiÚjÚx_embÚy_embre   s
             r"   r0   z PositionEmbeddingLearned.forwardÏ   sý   € ØÔˆØŒw�r˜�rŒ{‰ˆˆ1ÝŒL˜ 1¤8Ð,Ñ,Ô,ˆÝŒL˜ 1¤8Ð,Ñ,Ô,ˆØ—’˜qÑ!Ô!ˆØ—’˜qÑ!Ô!ˆåŒIà—O’O AÑ&Ô&×-Ò-¨a°°AÑ6Ô6Ø—O’O AÑ&Ô&×-Ò-¨a°°AÑ6Ô6ðð ðñ ô ÷ ŠY�q‰\Œ\ßŠV�A�q˜!œ' !œ* aÑ(Ô(ð 	ð ˆ
r#   )rp   )r   r   r   r   rg   )r5   r   r   r   )	rh   ri   rj   rk   r   r4   rv   r0   rl   rm   s   @r"   ro   ro   ¾   s„   ø€ € € € € Ø*Ð*ðð ð ð ð ð ð ð"ð "ð "ð "ð0ð 0ð 0ð 0ðð ð ð ð ð ð ð r#   ro   Ú
hidden_dimr   Úposition_embeddingÚstrr   ú	nn.Modulec                óŠ   — | dz  }|dv rt          |d¬¦  «        S |dv rt          d|›d�¦  «        ‚t          d|›d�¦  «        ‚)	u7  Build a positional encoding module.

    Args:
        hidden_dim: Transformer hidden dimension. Half of this value is used as the number
            of positional feature dimensions for the sine encoding.
        position_embedding: Encoding variant to construct.  Supported values:
            ``"sine"`` / ``"v2"`` â€” standard sine/cosine positional encoding (recommended).
            The aliases ``"learned"`` / ``"v3"`` are **not supported** and raise
            :exc:`ValueError`; their implementation had two bugs (wrong ``forward`` signature
            and wrong shape unpacking) and are rejected early rather than silently producing
            incorrect results.

    Returns:
        Positional encoding module.

    Raises:
        ValueError: If *position_embedding* is not a recognised and supported variant.

    Examples:
        >>> import torch
        >>> from rfdetr.models.position_encoding import build_position_encoding
        >>> enc = build_position_encoding(256, "sine")
        >>> enc  # doctest: +ELLIPSIS
        PositionEmbeddingSine(...)
    r   )Úv2ÚsineT)r   )Úv3Úlearnedzposition_embedding=a   is not supported. The 'learned'/'v3' implementation has two bugs: the forward() signature is incompatible with Joiner.forward(), and the shape unpacking uses x.shape[:2] (batch, channels) instead of x.shape[-2:] (height, width). Use 'sine' or 'v2' instead.z2 is not supported. Supported values: 'sine', 'v2'.)r   r   )r†   r‡   Ú	num_stepss      r"   Úbuild_position_encodingr�   å   s}   € ð4 ˜a‘€IØ˜^Ð+Ð+å$ Y¸$Ð?Ñ?Ô?Ð?ØÐ.Ð.Ð.Ýð*Ð"4ð *ð *ð *ñ
ô 
ð 	
õ ÐsÐ+=ÐsÐsÐsÑ
tÔ
tÐtr#   )r†   r   r‡   rˆ   r   r‰   )rk   Ú
__future__r   r   Úcollections.abcr   Útypingr   rT   r   r   Úrfdetr.utilities.tensorsr   rB   ÚModuler   ro   r�   © r#   r"   ú<module>r—      s  ðð 8Ð 7à "Ð "Ð "Ð "Ð "Ð "à €€€Ø $Ð $Ð $Ð $Ð $Ð $Ø Ð Ð Ð Ð Ð à €€€Ø Ð Ð Ð Ð Ð Ð Ð à 1Ð 1Ð 1Ð 1Ð 1Ð 1ð Ð ðZð Zð Zð Zð Z˜BœIñ Zô Zð Zðz$ð $ð $ð $ð $˜rœyñ $ô $ð $ðN&uð &uð &uð &uð &uð &ur#   