§
    l¬¸jâ  ã                  óÊ   — d Z ddlmZ ddlZddlZddlZddlmZmZ  G d„ dej        j	        ¦  «        Z
 G d„ dej	        ¦  «        Z G d	„ d
ej	        ¦  «        Zd„ Zddd„ZdS )zVarious utility models.é    )ÚannotationsN)ÚTensorÚnnc                  óF   ‡ — e Zd ZdZ	 	 	 dˆ fd„	Zed„ ¦   «         Zd„ Zˆ xZS )	ÚDotProductScoringz^A module that computes dot-product scores between query features and pooled prompt embeddings.NTç      (@c                ó¾  •— t          ¦   «                              ¦   «          || _        t          |t          j        j        ¦  «        s|�J ‚|| _        t          j                             ||¦  «        | _	        t          j                             ||¦  «        | _
        t          dt          j        |¦  «        z  ¦  «        | _        || _        | j        r	|| _        dS dS )z(Initialize the DotProductScoring module.Nç      ð?)ÚsuperÚ__init__Úd_projÚ
isinstanceÚtorchr   ÚModuleÚ
prompt_mlpÚLinearÚprompt_projÚhs_projÚfloatÚnpÚsqrtÚscaleÚclamp_logitsÚclamp_max_val)ÚselfÚd_modelr   r   r   r   Ú	__class__s         €úv/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/ultralytics/models/sam/sam3/model_misc.pyr   zDotProductScoring.__init__   s¾   ø€ õ 	‰Œ×ÒÑÔÐØˆŒÝ˜*¥e¤h¤oÑ6Ô6ÐL¸*Ð:LÐ:LÐLØ$ˆŒÝ œ8Ÿ?š?¨7°FÑ;Ô;ˆÔÝ”x—’ w°Ñ7Ô7ˆŒÝ˜3¥¤¨¡¤Ñ0Ñ1Ô1ˆŒ
Ø(ˆÔØÔð 	/Ø!.ˆDÔÐÐð	/ð 	/ó    c                óü   — |                       | j        ¦  «                             dd¦  «        d         }t          j        t          j        |d¬¦  «        d¬¦  «        }| |z                       d¬¦  «        |z  }|S )z;Mean-pool the prompt embeddings over the valid tokens only.é   r   ).N©Údimr
   )Úmin)ÚtoÚdtypeÚpermuter   ÚclampÚsum)ÚpromptÚprompt_maskÚis_validÚ	num_validÚpooled_prompts        r   Úmean_pool_textz DotProductScoring.mean_pool_text'   sy   € ð !�L×$Ò$ V¤\Ñ2Ô2×:Ò:¸1¸aÑ@Ô@ÀÔKˆå”K¥¤	¨(¸Ð :Ñ :Ô :ÀÐDÑDÔDˆ	à (Ñ*×/Ò/°AÐ/Ñ6Ô6¸ÑBˆØÐr   c                ó8  — |                      ¦   «         dk    r0|                      ¦   «         dk    r|                      ¦   «         dk    sJ ‚| j        �-|                      |                     |j        ¦  «        ¦  «        }|                      ||¦  «        }|                      |¦  «        }|                      |¦  «        }t          j        || 	                    d¦  «        ¦  «        }|| j
        z  }| j        r"|                     | j         | j        ¬¦  «         |S )z1Compute dot-product scores between hs and prompt.é   é   é   Néÿÿÿÿ)r$   Úmax)r#   r   r%   r&   r/   r   r   r   ÚmatmulÚ	unsqueezer   r   Úclamp_r   )r   Úhsr*   r+   r.   Úproj_pooled_promptÚproj_hsÚscoress           r   ÚforwardzDotProductScoring.forward2   s  € ð
 �vŠv‰xŒx˜1Š}ˆ} §¢¡¤°Ò!2Ð!2°{·²Ñ7HÔ7HÈAÒ7MÐ7MÐ7MÐMð Œ?Ð&Ø—_’_ V§Y¢Y¨r¬xÑ%8Ô%8Ñ9Ô9ˆFð ×+Ò+¨F°KÑ@Ô@ˆð "×-Ò-¨mÑ<Ô<ÐØ—,’,˜rÑ"Ô"ˆõ ”˜gÐ'9×'CÒ'CÀBÑ'GÔ'GÑHÔHˆØ�$”*Ñˆð Ôð 	KØ�MŠM˜tÔ1Ð1°tÔ7IˆMÑJÔJÐJàˆr   )NTr   )	Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   Ústaticmethodr/   r=   Ú__classcell__©r   s   @r   r   r      sx   ø€ € € € € ØhÐhð ØØð/ð /ð /ð /ð /ð /ð( ðð ñ „\ððð ð ð ð ð ð r   r   c                  ó2   ‡ — e Zd ZdZ	 	 ddˆ fd„Zdd„Zˆ xZS )Ú
LayerScalez;LayerScale module for per-channel scaling of layer outputs.çñhãˆµøä>Fr#   ÚintÚinit_valuesúfloat | TensorÚinplaceÚboolÚreturnÚNonec                ó²   •— t          ¦   «                              ¦   «          || _        t          j        |t          j        |¦  «        z  ¦  «        | _        dS )z!Initialize the LayerScale module.N)r   r   rK   r   Ú	Parameterr   ÚonesÚgamma)r   r#   rI   rK   r   s       €r   r   zLayerScale.__init__R   sD   ø€ õ 	‰Œ×ÒÑÔÐØˆŒÝ”\ +µ´
¸3±´Ñ"?Ñ@Ô@ˆŒ
ˆ
ˆ
r   Úxr   c                óX   — | j         r|                     | j        ¦  «        n	|| j        z  S )z%Apply LayerScale to the input tensor.)rK   Úmul_rR   )r   rS   s     r   r=   zLayerScale.forward]   s(   € à%)¤\ÐEˆq�vŠv�d”jÑ!Ô!Ð!°q¸4¼:±~ÐEr   )rG   F)r#   rH   rI   rJ   rK   rL   rM   rN   )rS   r   rM   r   )r>   r?   r@   rA   r   r=   rC   rD   s   @r   rF   rF   O   ss   ø€ € € € € ØEÐEð
 '+Øð		Að 	Að 	Að 	Að 	Að 	Að 	AðFð Fð Fð Fð Fð Fð Fð Fr   rF   c                  ó0   ‡ — e Zd ZdZ	 	 dd	ˆ fd„Zd„ Zˆ xZS )
ÚTransformerWrapperzEA wrapper for the transformer consisting of an encoder and a decoder.ÚnoneTr   rH   c                óú   •— t          ¦   «                              ¦   «          || _        || _        |�|j        nd| _        || _        |dv sJ d|› d�¦   «         ‚|| _        |                      ¦   «          || _        dS )z"Initialize the TransformerWrapper.N)rX   zunknown param z of two_stage_type)	r   r   ÚencoderÚdecoderÚnum_queriesÚpos_enc_at_input_decÚtwo_stage_typeÚ_reset_parametersr   )r   rZ   r[   r   r^   r]   r   s         €r   r   zTransformerWrapper.__init__e   s’   ø€ õ 	‰Œ×ÒÑÔÐØˆŒØˆŒØ29Ð2E˜7Ô.Ð.È4ˆÔØ$8ˆÔ!ð  Ð)Ð)Ð)Ð+^¸NÐ+^Ð+^Ð+^Ñ)Ô)Ð)Ø,ˆÔà×ÒÑ Ô Ð ØˆŒˆˆr   c                óÀ   — |                       ¦   «         D ]H\  }}|                     ¦   «         dk    r+d|vr'd|vr#d|vrt          j                             |¦  «         ŒIdS )z'Initialize the parameters of the model.r!   Ú	box_embedÚquery_embedÚreference_pointsN)Únamed_parametersr#   r   ÚinitÚxavier_uniform_)r   ÚnÚps      r   r_   z$TransformerWrapper._reset_parameters{   ss   € à×)Ò)Ñ+Ô+ð 	+ð 	+‰DˆAˆqØ�uŠu‰wŒw˜Š{ˆ{˜{°!Ð3Ð3¸ÈQÐ8NÐ8NÐSeÐmnÐSnÐSnÝ”×'Ò'¨Ñ*Ô*Ð*øð	+ð 	+r   )rX   T)r   rH   )r>   r?   r@   rA   r   r_   rC   rD   s   @r   rW   rW   b   s_   ø€ € € € € ØOÐOð Ø!ðð ð ð ð ð ð ð,+ð +ð +ð +ð +ð +ð +r   rW   c                ó2  — | j         \  }}}t          j        | dd…dd…df          d¦  «        }t          j        | dd…ddd…f          d¦  «        }|                     ¦   «         |z  }|                     ¦   «         |z  }t          j        ||gd¦  «        }|S )z:Compute the valid ratio of height and width from the mask.Nr   r!   r4   )Úshaper   r)   r   Ústack)	ÚmaskÚ_ÚHÚWÚvalid_HÚvalid_WÚvalid_ratio_hÚvalid_ratio_wÚvalid_ratios	            r   Úget_valid_ratioru   ‚   s™   € àŒj�G€A€qˆ!ÝŒi˜˜a˜a˜a    A˜gœ˜¨Ñ*Ô*€GÝŒi˜˜a˜a˜a  A A A˜gœ˜¨Ñ*Ô*€GØ—M’M‘O”O aÑ'€MØ—M’M‘O”O aÑ'€MÝ”+˜}¨mÐ<¸bÑAÔA€KØÐr   é   Ú
pos_tensorútorch.TensorÚ	num_featsrH   c           	     óx  — |dz  dk    sJ ‚|dz  }dt           j        z  }t          j        || j        | j        ¬¦  «        }ddt          j        |dd¬¦  «        z  |z  z  }| dd…dd…df         |z  }| dd…dd…df         |z  }|dd…dd…df         |z  }|dd…dd…df         |z  }t          j        |dd…dd…ddd…f                              ¦   «         |dd…dd…ddd…f          	                    ¦   «         fd	¬
¦  «         
                    d¦  «        }t          j        |dd…dd…ddd…f                              ¦   «         |dd…dd…ddd…f          	                    ¦   «         fd	¬
¦  «         
                    d¦  «        }|                      d¦  «        dk    rt          j        ||fd¬
¦  «        }�n€|                      d¦  «        dk    �rA| dd…dd…df         |z  }	|	dd…dd…df         |z  }
t          j        |
dd…dd…ddd…f                              ¦   «         |
dd…dd…ddd…f          	                    ¦   «         fd	¬
¦  «         
                    d¦  «        }
| dd…dd…d	f         |z  }|dd…dd…df         |z  }t          j        |dd…dd…ddd…f                              ¦   «         |dd…dd…ddd…f          	                    ¦   «         fd	¬
¦  «         
                    d¦  «        }t          j        |||
|fd¬
¦  «        }n%t          d|                      d¦  «        › �¦  «        ‚|S )a�  Generate sinusoidal position embeddings for 2D or 4D coordinate tensors.

    This function creates sinusoidal embeddings using sine and cosine functions at different frequencies, similar to the
    positional encoding used in Transformer models. It supports both 2D position tensors (x, y) and 4D tensors (x, y, w,
    h) for bounding box coordinates.

    Args:
        pos_tensor (torch.Tensor): Input position tensor of shape (n_query, bs, 2) for 2D coordinates or (n_query, bs,
            4) for 4D coordinates (bounding boxes).
        num_feats (int): Number of feature dimensions for the output embedding. Must be even. Defaults to 256.

    Returns:
        (torch.Tensor): Sinusoidal position embeddings of shape (n_query, bs, num_feats) for 2D input or (n_query, bs,
            num_feats * 2) for 4D input.

    Raises:
        AssertionError: If num_feats is not even.
        ValueError: If pos_tensor.size(-1) is not 2 or 4.

    Examples:
        >>> pos_2d = torch.rand(100, 8, 2)  # 100 queries, batch size 8, 2D coordinates
        >>> embeddings_2d = gen_sineembed_for_position(pos_2d, num_feats=256)
        >>> embeddings_2d.shape
        torch.Size([100, 8, 256])
        >>> pos_4d = torch.rand(50, 4, 4)  # 50 queries, batch size 4, 4D coordinates
        >>> embeddings_4d = gen_sineembed_for_position(pos_4d, num_feats=128)
        >>> embeddings_4d.shape
        torch.Size([50, 4, 256])
    r3   r   )r&   Údevicei'  Úfloor)Úrounding_modeNr!   r2   r"   r4   r1   zUnknown pos_tensor shape(-1):)ÚmathÚpir   Úaranger&   r{   Údivrk   ÚsinÚcosÚflattenÚsizeÚcatÚ
ValueError)rw   ry   r   Údim_tÚx_embedÚy_embedÚpos_xÚpos_yÚposÚw_embedÚpos_wÚh_embedÚpos_hs                r   Úgen_sineembed_for_positionr’   �   s³  € ð< �q‰=˜AÒÐÐÐØ˜Q‘€Ið •”‰K€EÝŒL˜¨*Ô*:À:ÔCTÐUÑUÔU€EØ�a�5œ9 U¨A¸WÐEÑEÔEÑFÈÑRÑS€EØ˜˜˜˜A˜A˜A˜q˜Ô! EÑ)€GØ˜˜˜˜A˜A˜A˜q˜Ô! EÑ)€GØ�A�A�A�q�q�q˜$�JÔ %Ñ'€EØ�A�A�A�q�q�q˜$�JÔ %Ñ'€EÝŒK˜˜q˜q˜q ! ! ! Q T¨ T˜zÔ*×.Ò.Ñ0Ô0°%¸¸¸¸1¸1¸1¸a¸dÀ¸d¸
Ô2C×2GÒ2GÑ2IÔ2IÐJÐPQÐRÑRÔR×ZÒZÐ[\Ñ]Ô]€EÝŒK˜˜q˜q˜q ! ! ! Q T¨ T˜zÔ*×.Ò.Ñ0Ô0°%¸¸¸¸1¸1¸1¸a¸dÀ¸d¸
Ô2C×2GÒ2GÑ2IÔ2IÐJÐPQÐRÑRÔR×ZÒZÐ[\Ñ]Ô]€EØ‡‚�rÑÔ˜aÒÐÝŒi˜ ˜¨AÐ.Ñ.Ô.ˆ‰Ø	�Š˜Ñ	Ô	 Ò	!Ñ	!Ø˜Q˜Q˜Q    1˜WÔ%¨Ñ-ˆØ˜˜˜˜1˜1˜1˜d˜
Ô# eÑ+ˆÝ”˜U 1 1 1 a a a¨¨¨A¨ :Ô.×2Ò2Ñ4Ô4°e¸A¸A¸A¸q¸q¸qÀ!À$ÀQÀ$¸JÔ6G×6KÒ6KÑ6MÔ6MÐNÐTUÐVÑVÔV×^Ò^Ð_`ÑaÔaˆà˜Q˜Q˜Q    1˜WÔ%¨Ñ-ˆØ˜˜˜˜1˜1˜1˜d˜
Ô# eÑ+ˆÝ”˜U 1 1 1 a a a¨¨¨A¨ :Ô.×2Ò2Ñ4Ô4°e¸A¸A¸A¸q¸q¸qÀ!À$ÀQÀ$¸JÔ6G×6KÒ6KÑ6MÔ6MÐNÐTUÐVÑVÔV×^Ò^Ð_`ÑaÔaˆåŒi˜  u¨eÐ4¸!Ð<Ñ<Ô<ˆˆåÐN¸¿ºÈÑ9LÔ9LÐNÐNÑOÔOÐOØ€Jr   )rv   )rw   rx   ry   rH   )rA   Ú
__future__r   r~   Únumpyr   r   r   r   r   r   rF   rW   ru   r’   © r   r   ú<module>r–      s  ðð
 Ð à "Ð "Ð "Ð "Ð "Ð "à €€€à Ð Ð Ð Ø €€€Ø Ð Ð Ð Ð Ð Ð Ð ð<ð <ð <ð <ð <˜œœñ <ô <ð <ð~Fð Fð Fð Fð F�”ñ Fô Fð Fð&+ð +ð +ð +ð +˜œñ +ô +ð +ð@ð ð ð9ð 9ð 9ð 9ð 9ð 9ð 9r   