§
    ­¸j,1  ã                  óV  — U d Z ddlmZ ddlmZmZ ddlmZ ddlZddl	m
c mZ ddlmZm
Z
  G d„ de
j        ¦  «        Zd(d„Zd„ d„ d„ d„ d„ d„ dœZded<   d)d*d„Z G d „ d!e
j        ¦  «        Z G d"„ d#e
j        ¦  «        Z G d$„ d%e
j        ¦  «        Z G d&„ d'e
j        ¦  «        ZdS )+z
Projector.é    )Úannotations)ÚCallableÚSequence)ÚcastN)ÚTensorÚnnc                  ó.   ‡ — e Zd ZdZddˆ fd	„Zdd„Zˆ xZS )Ú	LayerNormaH  A LayerNorm variant, popularized by Transformers, that performs point-wise mean and variance normalization over
    the channel dimension for inputs that have shape (batch_size, channels, height, width).

    https://github.com/facebookresearch/ConvNeXt/blob/d1fa8f6fef0a165b27399986cc2bdacc92777e40/models/convnext.py#L119
    ç�íµ ÷Æ°>Únormalized_shapeÚintÚepsÚfloatÚreturnÚNonec                ó  •— t          ¦   «                              ¦   «          t          j        t	          j        |¦  «        ¦  «        | _        t          j        t	          j        |¦  «        ¦  «        | _        || _	        |f| _
        d S ©N)ÚsuperÚ__init__r   Ú	ParameterÚtorchÚonesÚweightÚzerosÚbiasr   r   )Úselfr   r   Ú	__class__s      €úp/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/rfdetr/models/backbone/projector.pyr   zLayerNorm.__init__   si   ø€ Ý‰Œ×ÒÑÔÐÝ”l¥5¤:Ð.>Ñ#?Ô#?Ñ@Ô@ˆŒÝ”L¥¤Ð-=Ñ!>Ô!>Ñ?Ô?ˆŒ	ØˆŒØ!1Ð 3ˆÔÐÐó    Úxr   c                ó¾   — |                      dddd¦  «        }t          j        || j        | j        | j        | j        ¦  «        }|                      dddd¦  «        }|S )zb
        LayerNorm forward
        TODO: this is a hack to avoid overflow when using fp16
        r   é   é   é   )ÚpermuteÚFÚ
layer_normr   r   r   r   ©r   r    s     r   ÚforwardzLayerNorm.forward&   sW   € ð
 �IŠI�a˜˜A˜qÑ!Ô!ˆÝŒL˜˜DÔ1°4´;ÀÄ	È4Ì8ÑTÔTˆØ�IŠI�a˜˜A˜qÑ!Ô!ˆØˆr   )r   )r   r   r   r   r   r   ©r    r   r   r   ©Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   r)   Ú__classcell__©r   s   @r   r
   r
      s`   ø€ € € € € ðð ð4ð 4ð 4ð 4ð 4ð 4ð 4ðð ð ð ð ð ð ð r   r
   Únormú'str | Callable[[int], nn.Module] | NoneÚout_channelsr   r   únn.Module | Nonec                óŠ   — | €dS t          | t          ¦  «        r t          | ¦  «        dk    rdS dd„ i|          }  | |¦  «        S )zÞ
    Args:
        norm: Either one of BN, SyncBN, FrozenBN, GN;
            or a callable that takes a channel number and returns the normalization layer as a nn.Module.

    Returns:
        The normalization layer.
    Nr   ÚLNc                ó    — t          | ¦  «        S r   )r
   )Úchannelss    r   ú<lambda>zget_norm.<locals>.<lambda>@   s   € ¥9¨XÑ#6Ô#6€ r   )Ú
isinstanceÚstrÚlen)r2   r4   s     r   Úget_normr>   1   s^   € ð €|ØˆtÝ�$�ÑÔð Ýˆt‰9Œ9˜Š>ˆ>Ø�4àÐ6Ð6ð
à
ôˆð ˆ4�ÑÔÐr   c                ó,   — t          j        | ¬¦  «        S ©N©Úinplace)r   ÚSiLUrA   s    r   r:   r:   G   ó   € �BœG¨GÐ4Ñ4Ô4€ r   c                ó,   — t          j        | ¬¦  «        S r@   )r   ÚReLUrA   s    r   r:   r:   H   rD   r   c                ó.   — t          j        d| ¬¦  «        S ©Ngš™™™™™¹?rA   ©r   Ú	LeakyReLUrA   s    r   r:   r:   I   ó   € ¥¤¨c¸7Ð!CÑ!CÔ!C€ r   c                ó.   — t          j        d| ¬¦  «        S rH   rI   rA   s    r   r:   r:   J   rK   r   c                ó.   — t          j        d| ¬¦  «        S rH   rI   rA   s    r   r:   r:   K   s   € �Rœ\¨#°wÐ?Ñ?Ô?€ r   c                ó(   — t          j        ¦   «         S r   )r   ÚIdentityrA   s    r   r:   r:   L   s   € �"œ+™-œ-€ r   )ÚsiluÚrelurJ   Ú	leakyreluÚlreluNz-dict[str | None, Callable[[bool], nn.Module]]Ú_ACTIVATIONSFÚnameú
str | NonerB   Úboolú	nn.Modulec                óp   — 	 t          |          |¦  «        S # t          $ r t          d| › �¦  «        d‚w xY w)zGet activation.zUnsupported act type: N)rT   ÚKeyErrorÚAttributeError)rU   rB   s     r   Úget_activationr\   P   sS   € ðHÝ˜DÔ! 'Ñ*Ô*Ð*øÝð Hð Hð HÝÐ<°dÐ<Ð<Ñ=Ô=À4ÐGðHøøøs   ‚ —5c                  ó<   ‡ — e Zd ZdZ	 	 	 	 	 	 	 ddˆ fd„Zdd„Zˆ xZS )ÚConvXzConv-bn module.r#   r$   rQ   FÚ	in_planesr   Ú
out_planesÚkernelúint | tuple[int, int]ÚstrideÚgroupsÚdilationÚactr<   r'   rW   Úrms_normr   r   c
           
     óÖ  •— t          ¦   «                              ¦   «          t          |t          ¦  «        s||f}|d         dz  |d         dz  f}
t	          j        |||||
||d¬¦  «        | _        |  |	rt	          j        |¦  «        | _        nC|r(t          t          j
        t          d|¦  «        ¦  «        nt	          j        |¦  «        | _        t          |d¬¦  «        | _        d S )	Nr   r"   r$   F)Úkernel_sizerc   Úpaddingrd   re   r   r7   TrA   )r   r   r;   Útupler   ÚConv2dÚconvÚRMSNormÚbnr   ÚModuler>   ÚBatchNorm2dr\   rf   )r   r_   r`   ra   rc   rd   re   rf   r'   rg   rj   r   s              €r   r   zConvX.__init__[   sñ   ø€ õ 	‰Œ×ÒÑÔÐÝ˜&¥%Ñ(Ô(ð 	&Ø˜fÐ%ˆFØ˜!”9 ‘> 6¨!¤9°¡>Ð2ˆÝ”IØØØØØØØØð	
ñ 	
ô 	
ˆŒ	ð 	ÐØð 	pÝ”j Ñ,Ô,ˆDŒGˆGàEOÐo•d�2œ9¥h¨t°ZÑ&@Ô&@ÑAÔAÐAÕUWÔUcÐdnÑUoÔUoˆDŒGÝ! #¨tÐ4Ñ4Ô4ˆŒˆˆr   r    r   c                óÆ   — |                       |                      |                      |                     ¦   «         ¦  «        ¦  «        ¦  «        }t	          t
          |¦  «        S )zforward.)rf   ro   rm   Ú
contiguousr   r   )r   r    Úouts      r   r)   zConvX.forward|   sC   € à�hŠh�t—w’w˜tŸyšy¨¯ª©¬Ñ8Ô8Ñ9Ô9Ñ:Ô:ˆÝ•F˜CÑ Ô Ð r   )r#   r$   r$   r$   rQ   FF)r_   r   r`   r   ra   rb   rc   r   rd   r   re   r   rf   r<   r'   rW   rg   rW   r   r   r*   r+   r1   s   @r   r^   r^   X   st   ø€ € € € € ØÐð )*ØØØØØ Øð5ð 5ð 5ð 5ð 5ð 5ð 5ðB!ð !ð !ð !ð !ð !ð !ð !r   r^   c                  ó<   ‡ — e Zd ZdZ	 	 	 	 	 	 	 ddˆ fd„Zdd„Zˆ xZS )Ú
BottleneckzStandard bottleneck.Tr$   ©r#   r#   ç      à?rP   FÚc1r   Úc2ÚshortcutrW   ÚgÚkútuple[int, int]Úer   rf   r<   r'   rg   r   r   c
           
     ó  •— t          ¦   «                              ¦   «          t          ||z  ¦  «        }
t          ||
|d         d|||	¬¦  «        | _        t          |
||d         d||||	¬¦  «        | _        |o||k    | _        dS )z1ch_in, ch_out, shortcut, groups, kernels, expand.r   r$   ©rf   r'   rg   )rd   rf   r'   rg   N)r   r   r   r^   Úcv1Úcv2Úadd)r   ry   rz   r{   r|   r}   r   rf   r'   rg   Úc_r   s              €r   r   zBottleneck.__init__…   s†   ø€ õ 	‰Œ×ÒÑÔÐÝ��a‘‰[Œ[ˆÝ˜˜R  1¤ q¨c¸jÐS[Ð\Ñ\Ô\ˆŒÝ˜˜R  1¤ q°¸È
Ð]eÐfÑfÔfˆŒØÐ(  b¢ˆŒˆˆr   r    r   c           	     óÜ   — t          t          | j        r+||                      |                      |¦  «        ¦  «        z   n'|                      |                      |¦  «        ¦  «        ¦  «        S )z1'forward()' applies the YOLOv5 FPN to input data.)r   r   r„   rƒ   r‚   r(   s     r   r)   zBottleneck.forward˜   sU   € å•F¸¼Ð\˜A §¢¨¯ª°!©¬Ñ 5Ô 5Ñ5Ð5ÀtÇxÂxÐPT×PXÒPXÐYZÑP[ÔP[ÑG\ÔG\Ñ]Ô]Ð]r   )Tr$   rw   rx   rP   FF)ry   r   rz   r   r{   rW   r|   r   r}   r~   r   r   rf   r<   r'   rW   rg   rW   r   r   r*   r+   r1   s   @r   rv   rv   ‚   s{   ø€ € € € € ØÐð ØØ#ØØØ Øð)ð )ð )ð )ð )ð )ð )ð&^ð ^ð ^ð ^ð ^ð ^ð ^ð ^r   rv   c                  ó<   ‡ — e Zd ZdZ	 	 	 	 	 	 	 ddˆ fd„Zdd„Zˆ xZS )ÚC2fz<Faster Implementation of CSP Bottleneck with 2 convolutions.r$   Frx   rP   ry   r   rz   Únr{   rW   r|   r   r   rf   r<   r'   rg   r   r   c
           	     ó€  •‡ ‡‡‡‡‡	— t          ¦   «                              ¦   «          t          ||z  ¦  «        ‰ _        t	          |d‰ j        z  dd‰‰‰	¬¦  «        ‰ _        t	          d|z   ‰ j        z  |d‰‰‰	¬¦  «        ‰ _        t          j        ˆˆˆˆ	ˆ ˆfd„t          |¦  «        D ¦   «         ¦  «        ‰ _
        dS )z3ch_in, ch_out, number, shortcut, groups, expansion.r"   r$   r�   c              3  ó\   •K  — | ]&}t          ‰j        ‰j        ‰‰d d‰‰‰¬¦	  «	        V — Œ'dS )rw   ç      ð?)r}   r   rf   r'   rg   N)rv   Úc)Ú.0Ú_rf   r|   r'   rg   r   r{   s     €€€€€€r   ú	<genexpr>zC2f.__init__.<locals>.<genexpr>³   sX   øè è € ð 
ð 
àõ �t”v˜tœv x°°fÀÈÐYcÐnvÐwÑwÔwð
ð 
ð 
ð 
ð 
ð 
r   N)r   r   r   r�   r^   r‚   rƒ   r   Ú
ModuleListÚrangeÚm)r   ry   rz   r‰   r{   r|   r   rf   r'   rg   r   s   `   `` ```€r   r   zC2f.__init__    sã   øøøøøøø€ õ 	‰Œ×ÒÑÔÐÝ�R˜!‘V‘”ˆŒÝ˜˜Q ¤™Z¨¨A°3À:ÐX`ÐaÑaÔaˆŒÝØ�‰U�d”fÑ˜b !¨¸Èhð
ñ 
ô 
ˆŒõ ”ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
å˜1‘X”Xð
ñ 
ô 
ñ 
ô 
ˆŒˆˆr   r    r   c           	     óJ  ‡— t          |                      |¦  «                             | j        | j        fd¦  «        ¦  «        Š‰                     ˆfd„| j        D ¦   «         ¦  «         t          t          |                      t          j
        ‰d¦  «        ¦  «        ¦  «        S )z.Forward pass using split() instead of chunk().r$   c              3  ó:   •K  — | ]} |‰d          ¦  «        V — ŒdS )éÿÿÿÿN© )rŽ   r“   Úys     €r   r�   zC2f.forward.<locals>.<genexpr>»   s/   øè è € Ð*Ð*˜a���1�R”5‘”Ð*Ð*Ð*Ð*Ð*Ð*r   )Úlistr‚   Úsplitr�   Úextendr“   r   r   rƒ   r   Úcat)r   r    r˜   s     @r   r)   zC2f.forward¸   s‚   ø€ å�—’˜!‘”×"Ò" D¤F¨D¬FÐ#3°QÑ7Ô7Ñ8Ô8ˆØ	�ŠÐ*Ð*Ð*Ð* 4¤6Ð*Ñ*Ô*Ñ*Ô*Ð*Ý•F˜DŸHšH¥U¤Y¨q°!¡_¤_Ñ5Ô5Ñ6Ô6Ð6r   )r$   Fr$   rx   rP   FF)ry   r   rz   r   r‰   r   r{   rW   r|   r   r   r   rf   r<   r'   rW   rg   rW   r   r   r*   r+   r1   s   @r   rˆ   rˆ   �   ss   ø€ € € € € ØFÐFð ØØØØØ Øð
ð 
ð 
ð 
ð 
ð 
ð 
ð07ð 7ð 7ð 7ð 7ð 7ð 7ð 7r   rˆ   c                  ó8   ‡ — e Zd ZdZ	 	 	 	 	 ddˆ fd„Zdd„Zˆ xZS )ÚMultiScaleProjectorz‹This module implements MultiScaleProjector in :paper:`lwdetr`.

    It creates pyramid features built on top of the input feature map.
    r#   FrŒ   r   Úin_channelsúSequence[int]r4   r   Úscale_factorsúSequence[float]Ú
num_blocksr'   rW   rg   Úsurvival_probr   Úforce_drop_last_n_featuresr   r   c	                óà  •‡— t          ¦   «                              ¦   «          || _        || _        || _        g }	g }
d| _        |D �]õŠ‰dk    rd| _        Œg }|D �]3}g }‰dk    r‰|                     t          j        ||dz  dd¬¦  «        t          t          j
        t          d|dz  ¦  «        ¦  «        t          j        ¦   «         t          j        |dz  |dz  dd¬¦  «        g¦  «         n~‰d	k    r0|                     t          j        ||dz  dd¬¦  «        g¦  «         nH‰d
k    rnA‰dk    r)|                     t          ||dd|¬¦  «        g¦  «         nt          d‰› �¦  «        ‚|                     t          j        |Ž ¦  «         �Œ5|	                     t          j        |¦  «        ¦  «         t%          t'          ˆfd„|D ¦   «         ¦  «        ¦  «        }t)          ||||¬¦  «        t          t          j
        t          d|¦  «        ¦  «        g}|
                     t          j        |Ž ¦  «         �Œ÷t          j        |	¦  «        | _        t          j        |
¦  «        | _        dS )a2  
        Args:
            in_channels: Channels in each input feature map level.
            out_channels: Number of channels in the output feature maps.
            scale_factors: List of scaling factors to upsample or downsample
                the input features for creating pyramid features.
        Fg      Ð?Tg      @r"   )ri   rc   r7   é   g       @rŒ   rx   r#   )r'   zUnsupported scale_factor:c              3  ó>   •K  — | ]}|t          d ‰¦  «        z  V — ŒdS )r$   N)Úmax)rŽ   Ú
in_channelÚscales     €r   r�   z/MultiScaleProjector.__init__.<locals>.<genexpr>  s0   øè è € ÐWÐW¸Z˜Z­3¨q°%©=¬=Ñ8ÐWÐWÐWÐWÐWÐWr   N)r   r   r¡   r¤   r¥   Úuse_extra_poolr›   r   ÚConvTranspose2dr   rp   r>   ÚGELUr^   ÚNotImplementedErrorÚappendÚ
Sequentialr‘   r   Úsumrˆ   Ústages_samplingÚstages)r   rŸ   r4   r¡   r£   r'   rg   r¤   r¥   r³   r´   Úscale_stage_layersÚin_dimÚlayersÚstage_layersr«   r   s                  @€r   r   zMultiScaleProjector.__init__Å   s¦  øø€ õ$ 	‰Œ×ÒÑÔÐà*ˆÔØ*ˆÔØ*DˆÔ'à/1ˆØ&(ˆà#ˆÔØ"ð :	8ñ :	8ˆEØ˜Š}ˆ}ð '+�Ô#Øà24ÐØ%ð *Bñ *B�Ø*,�ð ˜C’<�<Ø—M’MåÔ.¨v°vÀ±{ÐPQÐZ[Ð\Ñ\Ô\Ý ¥¤­H°T¸6ÀQ¹;Ñ,GÔ,GÑHÔHÝœG™IœIÝÔ.¨v¸©{¸FÀa¹KÐUVÐ_`ÐaÑaÔað	ñô ð ð ð ˜c’\�\ð —M’MåÔ.¨v°vÀ±{ÐPQÐZ[Ð\Ñ\Ô\ðñô ð ð ð ˜c’\�\ØØ˜c’\�\Ø—M’Må! &¨&°!°QÀ:ÐNÑNÔNðñô ð ð õ .Ð.QÈ%Ð.QÐ.QÑRÔRÐRØ"×)Ò)­"¬-¸Ð*@ÑAÔAÐAÑAØ×"Ò"¥2¤=Ð1CÑ#DÔ#DÑEÔEÐEå�ÐWÐWÐWÐWÈ;ÐWÑWÔWÑWÔWÑXÔXˆFå�F˜L¨*ÀÐLÑLÔLÝ•R”Y¥¨¨|Ñ <Ô <Ñ=Ô=ð-ˆLð �MŠM�"œ-¨Ð6Ñ7Ô7Ð7Ñ7å!œ}¨_Ñ=Ô=ˆÔÝ”m FÑ+Ô+ˆŒˆˆr   r    úlist[Tensor]c                óÚ  — t          |¦  «        }| j        dk     rˆ| j        r�t          |¦  «        }d| j        z
  }t	          j        d¦  «                             ¦   «         }t          d|¦  «        D ]0}|||dz
  z  z  }||k     rt	          j        ||         ¦  «        ||<   Œ1nG| j	        dk    r<t          | j	        ¦  «        D ]'}t	          j        ||dz             ¦  «        ||dz    <   Œ(g }t          | j        ¦  «        D ]²\  }}g }	t          t          t          j        | j        |         ¦  «        ¦  «        D ])\  }
}|	                      |||
         ¦  «        ¦  «         Œ*t          |	¦  «        dk    rt	          j        |	d¬¦  «        n|	d         }|                      ||¦  «        ¦  «         Œ³| j        r1|                     t'          j        |d         ddd¬¦  «        ¦  «         |S )	aÆ  
        Args:
            x: Tensor of shape (N,C,H,W). H, W must be a multiple of ``self.size_divisibility``.
        Returns:
            dict[str->Tensor]:
                mapping from feature map name to pyramid feature map tensor
                in high to low resolution order. Returned feature names follow the FPN
                convention: "p<stage>", where stage has stride = 2 ** stage e.g.,
                ["p2", "p3", ..., "p6"].
        rŒ   r$   r—   r   )Údimr–   r"   )ri   rc   rj   )r=   r¤   Útrainingr™   r   ÚrandÚitemr’   Ú
zeros_liker¥   Ú	enumerater´   r   r   r‘   r³   r°   rœ   r¬   r&   Ú
max_pool2d)r   r    Únum_featuresÚfinal_drop_probÚdrop_pÚiÚcritical_drop_probÚresultsÚstageÚfeat_fuse_listÚjÚstage_samplingÚ	feat_fuses                r   r)   zMultiScaleProjector.forward   s  € õ ˜1‘v”vˆØÔ Ò#Ð#¨¬Ð#Ý�Q‘”ˆAØ $Ô"4Ñ4ˆOå”Z ‘^”^×(Ò(Ñ*Ô*ˆFÝ˜1˜lÑ+Ô+ð 2ð 2�Ø%&¨/¸\ÈAÑ=MÑ*NÑ%OÐ"ØÐ.Ò.Ð.Ý Ô+¨A¨a¬DÑ1Ô1�A�a‘Døð2ð Ô,¨qÒ0Ð0Ý˜4Ô:Ñ;Ô;ð <ð <�å#Ô.¨q°1°q±5°¬{Ñ;Ô;��A˜‘E�(‘�àˆå! $¤+Ñ.Ô.ð 	-ð 	-‰HˆAˆuØˆNÝ%.­tµB´MÀ4ÔCWÐXYÔCZÑ/[Ô/[Ñ%\Ô%\ð <ð <Ñ!��>Ø×%Ò% n n°Q°q´TÑ&:Ô&:Ñ;Ô;Ð;Ð;Ý<?ÀÑ<OÔ<OÐRSÒ<SÐ<S�œ	 .°aÐ8Ñ8Ô8Ð8ÐYgÐhiÔYjˆIØ�NŠN˜5˜5 Ñ+Ô+Ñ,Ô,Ð,Ð,ØÔð 	ZØ�NŠN�1œ<¨°¬ÀÈ1ÐVWÐXÑXÔXÑYÔYÐYØˆr   )r#   FFrŒ   r   )rŸ   r    r4   r   r¡   r¢   r£   r   r'   rW   rg   rW   r¤   r   r¥   r   r   r   )r    r¹   r   r¹   r+   r1   s   @r   rž   rž   ¿   sy   ø€ € € € € ðð ð Ø ØØ"Ø*+ðY,ð Y,ð Y,ð Y,ð Y,ð Y,ð Y,ðv$ð $ð $ð $ð $ð $ð $ð $r   rž   )r2   r3   r4   r   r   r5   )F)rU   rV   rB   rW   r   rX   )r/   Ú
__future__r   Úcollections.abcr   r   Útypingr   r   Útorch.nn.functionalr   Ú
functionalr&   r   rp   r
   r>   rT   Ú__annotations__r\   r^   rv   rˆ   rž   r—   r   r   ú<module>rÓ      sú  ðð Ð Ð à "Ð "Ð "Ð "Ð "Ð "à .Ð .Ð .Ð .Ð .Ð .Ð .Ð .Ø Ð Ð Ð Ð Ð à €€€Ø Ð Ð Ð Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ð Ð ðð ð ð ð �”	ñ ô ð ð2ð ð ð ð, 5Ð4Ø4Ð4ØCÐCØCÐCØ?Ð?Ø
'Ð
'ð?ð ?€ð ð ð ñ ðHð Hð Hð Hð Hð'!ð '!ð '!ð '!ð '!ˆBŒIñ '!ô '!ð '!ðT^ð ^ð ^ð ^ð ^�”ñ ^ô ^ð ^ð67ð 7ð 7ð 7ð 7ˆ"Œ)ñ 7ô 7ð 7ðDEð Eð Eð Eð E˜"œ)ñ Eô Eð Eð Eð Er   