§
    �ßjã9  ã                  ó  — d Z ddlmZ ddlZddlZddlmZ ddlZddlZddl	Z	ddl
Zddl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mZ dd
lmZmZmZm Z m!Z!m"Z" dd„Z#d„ Z$d„ Z%d„ Z& G d„ de¦  «        Z'd„ Z(dS )aà  Pinned RF-DETR-Seg Small with an explicit padding-ignore research adapter.

The official PTL module/data/trainer API is used unmodified on disk. Its default
square stretch and padded-background mask costs are NOT appropriate for these
ROIs: inputs are pre-letterboxed and both matching and mask loss exclude padding.
Detection/classification/box losses, architecture, EMA and checkpoint selection
remain the upstream RF-DETR implementation. No pose or annotation metadata inputs.
é    )ÚannotationsN)ÚPath)Únn)Ú
functional)ÚDatasetÚSubseté   )ÚdigestÚdisk_budgetÚreadÚrequireÚunletterboxÚverify_datasetÚwrite)Ú
MAX_EPOCHSÚPATIENCEÚnative_metricsÚquality_reportÚresult_recordÚseed_allc           
     óø  — | d         }t          |t          j        ¦  «        r|€|n||         S |€+t          j        d|d         |d         ¦  «        |d         z   S |\  }}g }|                     d¬¦  «        D ]Z}|||k             }|                     t          j        d	|d         |         |d         ||f         ¦  «        |d         z   ¦  «         Œ[|rt          j        |¦  «        n|d         dd
…d
f         S )zGMaterialize matched queries only for loss, sample features for matcher.Ú
pred_masksNzbchw,bqc->bqhwÚspatial_featuresÚquery_featuresÚbiasT)Úsortedzchw,qc->qhwr   )Ú
isinstanceÚtorchÚTensorÚeinsumÚuniqueÚappendÚcat)ÚoutputsÚindicesÚvalueÚbatchÚqueryÚpartsÚindexÚchosens           úN/root/private/project/pallet_detection_v2/pallet/mask_model_training_rfdetr.pyÚprojected_mask_logitsr-       s*  € à�LÔ!€EÝ�%�œÑ&Ô&ð <Ø˜ˆuˆu¨U°7¬^Ð;Ø€ÝŒ|Ð,¨eÐ4FÔ.GÈÐO_ÔI`ÑaÔaÐbgÐhnÔboÑoÐoØ�L€Eˆ5Ø€EØ—’ T�Ñ*Ô*ð Yð YˆØ�u ’~Ô&ˆØ�Š•U”\ -°Ð7IÔ1JÈ5Ô1QØ"'Ð(8Ô"9¸%À¸-Ô"HñJô JØJOÐPVÌ-ñXñ 	Yô 	Yð 	Yð 	Yà$ÐJ�5Œ9�UÑÔÐ¨%Ð0BÔ*CÀBÀQÀBÈÀEÔ*JÐJó    c           	     óÄ  — ddl m} |d         }t          |t          j        ¦  «        r¶|j        dd…         }t          j        dt          d|d         |d         z  | j        z  ¦  «        d|j	        ¬	¦  «        } || 
                    dd¦  «        dd…df         |                     |j        d         |j        d         z  dd¦  «        d
¬¦  «        dd…df         }nÂ|d         |d         }	}|j        dd…         }t          j        dt          d|d         |d         z  | j        z  ¦  «        d|j	        ¬	¦  «        } |||                     |j        d         dd¦  «        d
¬¦  «        }
t          j        d|
|	¦  «        |d         z    
                    dd¦  «        }t          j        d„ |D ¦   «         ¦  «                             |j        ¦  «        }t          j        d„ |D ¦   «         ¦  «                             |j        ¦  «        } ||dd…df         |                     t!          |¦  «        dd¦  «        dd
¬¦  «        dd…df         } ||dd…df         |                     t!          |¦  «        dd¦  «        dd
¬¦  «        dd…df         }||z  d|z
  |z  }}t#          j        | ¦  «        |j        z  t#          j        |¦  «        |j        z  z   |                     d¦  «                             d¦  «        d         z  }|                     ¦   «         }dd||j        z  z  dz   ||j        z  |                     d¦  «        d         z   dz   z  z
  }||fS )zGSame point-sampled pairwise CE/Dice, with target-specific valid pixels.r   )Úpoint_sampler   éþÿÿÿNr	   é€   é   )ÚdeviceF)Úalign_cornersr   r   zbcp,bqc->bqpr   c                ó   — g | ]
}|d          ‘ŒS ©Úmasks© ©Ú.0Úts     r,   ú
<listcomp>z,padding_aware_mask_costs.<locals>.<listcomp>?   s   € Ð3Ð3Ð3 a�q˜”zÐ3Ð3Ð3r.   c                óz   — g | ]8}|d          d                               t          |d         ¦  «        dd¦  «        ‘Œ9S )Úpixel_validNr8   éÿÿÿÿ©ÚexpandÚlenr:   s     r,   r=   z,padding_aware_mask_costs.<locals>.<listcomp>@   s?   € ÐbÐbÐbÐUV˜!˜MÔ*¨4Ô0×7Ò7½¸A¸g¼J¹¼ÈÈRÑPÔPÐbÐbÐbr.   Únearest)Úmoder5   )Ú rfdetr.models.heads.segmentationr0   r   r   r   ÚshapeÚrandÚmaxÚmask_point_sample_ratior4   ÚflattenÚrepeatr    r#   ÚtoÚdtyperC   ÚFÚsoftplusÚTÚsumÚ	clamp_minÚsigmoid)Úselfr$   Útargetsr0   r&   ÚsizeÚcoordsÚlogitsÚfeaturesÚqueriesÚsamplesr8   ÚvalidityÚlabelÚvalidÚpositiveÚnegativeÚceÚprobabilityÚdices                       r,   Úpadding_aware_mask_costsre   0   sn  € à=Ð=Ð=Ð=Ð=Ð=Ø�LÔ!€EÝ�%�œÑ&Ô&ð 
^ØŒ{˜2˜3˜3ÔˆÝ”˜A�s 3¨¨Q¬°°Q´©¸Ô9UÑ(UÑVÔVÐXYÐbgÔbnÐoÑoÔoˆØ�˜eŸmšm¨A¨qÑ1Ô1°!°!°!°T°'Ô:Ø$Ÿmšm¨E¬K¸¬N¸5¼;Àq¼>Ñ,IÈ1ÈaÑPÔPÐ`eðgñ gô gØghÐghÐghÐjkÐgkômˆˆð "Ð"4Ô5°uÐ=MÔ7N�'ˆØŒ~˜b˜c˜cÔ"ˆÝ”˜A�s 3¨¨Q¬°°Q´©¸Ô9UÑ(UÑVÔVÐXYÐbjÔbqÐrÑrÔrˆØ�,˜x¨¯ª°x´~ÀaÔ7HÈ!ÈQÑ)OÔ)OÐ_dÐeÑeÔeˆÝ”,˜~¨w¸Ñ@Ô@ÀÀvÄÑN×WÒWÐXYÐ[\Ñ]Ô]ˆÝŒIÐ3Ð3¨7Ð3Ñ3Ô3Ñ4Ô4×7Ò7¸¼ÑEÔE€EÝŒyÐbÐbÐZaÐbÑbÔbÑcÔc×fÒfÐgmÔgsÑtÔt€HØˆL˜˜q˜q˜q $˜wœ¨¯ªµs¸5±z´zÀ1ÀaÑ)HÔ)HÈyÐhmÐnÑnÔnÐopÐopÐopÐrsÐosÔt€EØˆL˜ ! ! ! T 'Ô*¨F¯MªM½#¸e¹*¼*ÀaÈÑ,KÔ,KÐR[ÐkpÐqÑqÔqÐrsÐrsÐrsÐuvÐrvÔw€EØ˜u™ q¨¡w°¡oˆh€HÝ
Œ*�f�WÑ
Ô
 ¤
Ñ
*­Q¬Z¸Ñ-?Ô-?À(Ä*Ñ-LÑ
LÈeÏiÊiÐXYÉlÌl×NdÒNdÐefÑNgÔNgÐhlÔNmÑ	m€BØ—.’.Ñ"Ô"€KØˆa�˜xœzÑ)Ñ*¨1Ñ,¨{¸U¼WÑ/DÀXÇ\Â\ÐRSÁ_Ä_ÐUYÔEZÑ/ZÐ[\Ñ/\Ñ]Ñ]€DØˆtˆ8€Or.   c                óš  — t          j        |dd…df                              ¦   «         | j        dd…         d¬¦  «        dd…df         }t          j        |dd…df                              ¦   «         | j        dd…         d¬¦  «        dd…df         }t          j        | |d¬¦  «        |z                       d¦  «                             d¦  «        }||                     d¦  «                             d¦  «                             d¦  «        z  }|                      ¦   «         |z  }||z  }dd	||z                       d¦  «                             d¦  «        z  dz   ||z                        d¦  «                             d¦  «        dz   z  z
  }	t          |                     ¦   «         |z  |	                     ¦   «         |z  ¬
¦  «        S )zFDense matched-mask CE/Dice, nearest truth and explicit padding ignore.Nr1   rD   )rW   rE   r   Únone)Ú	reductionr	   r3   ©Úloss_mask_ceÚloss_mask_dice)
rO   ÚinterpolateÚfloatrG   Ú binary_cross_entropy_with_logitsrK   rR   rS   rT   Údict)
rY   rV   r_   Ú	num_boxesr^   Úweightrb   rc   r`   rd   s
             r,   Úmasked_dense_lossesrr   J   s¸  € õ ŒM˜' ! ! ! T 'Ô*×0Ò0Ñ2Ô2¸¼ÀbÀcÀcÔ9JÐQZÐ[Ñ[Ô[Ð\]Ð\]Ð\]Ð_`Ð\`Ôa€EÝŒ]˜5    D œ>×/Ò/Ñ1Ô1¸¼ÀRÀSÀSÔ8IÐPYÐZÑZÔZÐ[\Ð[\Ð[\Ð^_Ð[_Ô`€FÝ
Ô
,¨V°UÀfÐ
MÑ
MÔ
MÈfÑ
T×	]Ò	]Ð^_Ñ	`Ô	`×	dÒ	dÐefÑ	gÔ	g€BØ	ˆF�NŠN˜1ÑÔ×!Ò! !Ñ$Ô$×.Ò.¨qÑ1Ô1Ñ	1€BØ—.’.Ñ"Ô" 6Ñ)€KØ�V‰|€HØˆa�˜XÑ%×.Ò.¨qÑ1Ô1×5Ò5°aÑ8Ô8Ñ8¸Ñ:Ø	�XÑ	×&Ò& qÑ)Ô)×-Ò-¨aÑ0Ô0°Ñ2ñ4ñ 4€Då˜RŸVšV™XœX iÑ/ÀÇÂÁ
Ä
È9Ñ@TÐUÑUÔUÐUr.   c                ó˜  — |                       |¦  «        \  }}t          |||f¦  «        }t          |¦  «        s(|                     ¦   «         dz  }t	          ||¬¦  «        S t          j        d„ t          ||¦  «        D ¦   «         ¦  «        }	t          j        d„ t          ||¦  «        D ¦   «         ¦  «        }
t          ||	|
|¦  «        S )Nr   ri   c                ó6   — g | ]\  }\  }}|d          |         ‘ŒS r7   r9   ©r;   ÚtargetÚ_r+   s       r,   r=   z+padding_aware_mask_loss.<locals>.<listcomp>^   s*   € ÐaÐaÐaÑ4G°F¹K¸QÀ˜˜wœ¨Ô/ÐaÐaÐar.   c                óz   — g | ]8\  }\  }}|d          d                               t          |¦  «        dd¦  «        ‘Œ9S )r?   Nr@   rA   ru   s       r,   r=   z+padding_aware_mask_loss.<locals>.<listcomp>_   sX   € ð Ið Ið IÙ.˜6¡; A vð ˜mÔ,¨TÔ2×9Ò9½#¸f¹+¼+ÀrÈ2ÑNÔNð Ið Ið Ir.   )	Ú_get_src_permutation_idxr-   rC   rR   ro   r   r#   Úziprr   )rU   r$   rV   r%   rp   r'   r(   rY   ÚzeroÚlabelsr_   s              r,   Úpadding_aware_mask_lossr}   X   sÙ   € Ø×0Ò0°Ñ9Ô9�L€Eˆ5Ý" 7¨U°E¨NÑ;Ô;€FÝˆu‰:Œ:ð <Ø�zŠz‰|Œ|˜A‰~ˆÝ °dÐ;Ñ;Ô;Ð;ÝŒYÐaÐaÍ3ÈwÐX_ÑK`ÔK`ÐaÑaÔaÑbÔb€FÝŒIð Ið IÝ25°g¸wÑ2GÔ2GðIñ Iô Iñ Jô J€Eå˜v v¨u°iÑ@Ô@Ð@r.   c                  ó0   ‡ — e Zd Zˆ fd„Zd„ Zd„ Zd„ Zˆ xZS )ÚValidityDatasetc                óö   •— t          ¦   «                              |t          t          |¦  «        ¦  «        ¦  «         ||c| _        | _        t          |dd ¦  «        | _        t          |dd ¦  «        | _        d S )NÚcocoÚclass_names)	ÚsuperÚ__init__ÚrangerC   ÚbaseÚ
transformsÚgetattrr�   r‚   )rU   r†   r‡   Ú	__class__s      €r,   r„   zValidityDataset.__init__e   sh   ø€ å‰Œ×Ò˜�u¥S¨¡Y¤YÑ/Ô/Ñ0Ô0Ð0Ø%)¨:Ð"ˆŒ	�4”?Ý˜D &¨$Ñ/Ô/ˆŒ	Ý" 4¨¸Ñ=Ô=ˆÔÐÐr.   c                ó*   — t          | j        ¦  «        S ©N)rC   r†   )rU   s    r,   Ú__len__zValidityDataset.__len__l   s   € Ý�4”9‰~Œ~Ðr.   c                ó¾  — | j         |         \  }}t          t          |d                              ¦   «         ¦  «        ¦  «        }| j        |         }|d         \  }}t          t          |j        dd …         ¦  «        ||fk    d¦  «         |d         \  }}	|d         \  }
}t          j	        ||ft          j
        ¬¦  «        }d||	|	|z   …|||
z   …f<   ||d	<   ||fS )
NÚimage_idÚoutput_sizer1   z'RF-DETR changed letterboxed input shapeÚpadding_left_topÚscaled_size)rN   Tr?   )r†   ÚstrÚintÚitemr‡   r   ÚtuplerG   r   ÚzerosÚbool)rU   r*   Úimagerv   rŽ   Ú	transformÚwidthÚheightÚleftÚtopÚnwÚnhr_   s                r,   Ú__getitem__zValidityDataset.__getitem__o   sé   € Øœ	 %Ô(‰ˆˆvÝ•s˜6 *Ô-×2Ò2Ñ4Ô4Ñ5Ô5Ñ6Ô6ˆØ”O HÔ-ˆ	Ø! -Ô0‰ˆˆvÝ•�e”k " # #Ô&Ñ'Ô'¨F°E¨?Ò:Ð<eÑfÔfÐfØÐ0Ô1‰	ˆˆcØ˜=Ô)‰ˆˆBÝ”˜V U˜Oµ5´:Ð>Ñ>Ô>ˆØ*.ˆˆc�#�b‘&ˆj˜$˜t B™w˜,Ð&Ñ'Ø %ˆˆ}ÑØ�fˆ}Ðr.   c                ó    ‡ — ˆ fd„|D ¦   «         S )Nc                ó    •— g | ]
}‰|         ‘ŒS r9   r9   )r;   r*   rU   s     €r,   r=   z0ValidityDataset.__getitems__.<locals>.<listcomp>~   s   ø€ Ð1Ð1Ð1 ��U”Ð1Ð1Ð1r.   r9   )rU   r%   s   ` r,   Ú__getitems__zValidityDataset.__getitems__|   s   ø€ à1Ð1Ð1Ð1¨Ð1Ñ1Ô1Ð1r.   )Ú__name__Ú
__module__Ú__qualname__r„   rŒ   r    r£   Ú__classcell__)r‰   s   @r,   r   r   d   se   ø€ € € € € ð>ð >ð >ð >ð >ðð ð ðð ð ð2ð 2ð 2ð 2ð 2ð 2ð 2r.   r   c                óŒ  ‡ ‡&— dd l }ddlm} ddlm}m}m}	 t          ‰ dz  ¦  «        Š& G ˆ&fd„d|¦  «        }
 G ˆ fd„d|j        ¦  «        }g }d }t          d	¦  «        D �]é\  }}t          ¦   «          |                     d
d¬¦  «         |d|d›�z  }|                     d¬¦  «         |dz                       d¬¦  «          |dddt          |¦  «        dd¬¦  «        } |j        d`i dt          ‰ dz  ¦  «        “dt          |¦  «        “dt          “d|“dd|z  “dd“dd“d d!“d"d
“d#d“d$d“d%d“d&d“d'i “d(d“d)d“d*t           “d+d,“d-d“d.d“d/d“d0d“d1d “d2d3“d4d“d5d6g“d7d“d8d“d9d:“d;d“d<d“d=d>“d?d“Ž}|j        }t%          |d@z  t'          |                     dA¬¦  «        |                     dA¬¦  «        |dBdCdd3¬D¦  «        ¦  «         ~t+          j        ¦   «          t.          j                             ¦   «           |||¦  «        }t5          j        t8          |j        ¦  «        |j        _        t5          j        t>          |j        j         ¦  «        |j        j         _!         |
||¦  «        } |	||d:d¬E¦  «        }|j"         #                     ||¦  «        ¦  «         t.          j         $                    ¦   «          tK          j&        ¦   «         }	 | '                    ||¦  «         | #                    t'          |d|z  dFt.          j         (                    ¦   «         tK          j&        ¦   «         |z
  ¬G¦  «        ¦  «         |}|j)        }~~~t+          j        ¦   «          t.          j                             ¦   «           nÂ# t.          j        j*        $ r« | #                    t'          |d|z  dHt.          j         (                    ¦   «         tK          j&        ¦   «         |z
  ¬G¦  «        ¦  «         t%          |dIz  |dJ         ¦  «         ~~~t+          j        ¦   «          t.          j                             ¦   «          Y �Œçw xY wt%          |dKz  t'          ||r|dJ         d         nd dd¬L¦  «        ¦  «         tW          |d udM¦  «         |dNz  }tW          | ,                    ¦   «         dO¦  «         t[          ‰ ¦  «          |dddt          |¦  «        d¬P¦  «        }g }t          |¦  «        D �]\  }}|dQ         dRk    rŒt]          j/        t          ‰ dz  dSz  ta          |dT         ¦  «        j1        z  ¦  «        ¦  «        }tW          |d udU¦  «         t.          j         2                    ¦   «          tK          j&        ¦   «         }| 3                    |dVd¬W¦  «        }t.          j         2                    ¦   «          tK          j&        ¦   «         |z
  }t]          j/        t          ‰ |dX         z  ¦  «        t\          j4        ¦  «        dk    } tk          |¦  «        rœ|j6        �•to          tq          j9        |j:        ¦  «        ¦  «        }!tw          |j6        |!          <                    tp          j=        ¦  «        ‰&t          |dz   ¦  «                 ¦  «        }"|"dYk    }#t}          |j:        |!         ¦  «        }$ntq          j?        | ¦  «        dV}$}#| #                    t�          |tƒ          |#| ¦  «        |$|¦  «        ¦  «         �Œt…          ||¦  «        }%t%          |dZz  t'          t          | C                    |¦  «        ¦  «        t‰          |¦  «        d[dY|%d\         dd¬]¦  «        ¦  «         t'          |t          | C                    |¦  «        ¦  «        t‰          |¦  «        |%d^         |dJ         d         ¬_¦  «        S )aNr   )ÚRFDETRSegSmall)ÚRFDETRDataModuleÚRFDETRModelModuleÚbuild_trainerzprivate/rfdetr_transforms.jsonc                  ó"   •‡ — e Zd Zˆ ˆfd„Zˆ xZS )ú(train_rfdetr.<locals>.PadAwareDataModulec           	     ó  •— t          |dv d¦  «         t          ¦   «                              |¦  «         dD ]H}t          | |¦  «        }|�4t	          |t
          ¦  «        st          | |t          |‰¦  «        ¦  «         ŒId S )N)ÚfitÚvalidatez,Official val/test/predict datasets forbidden)Ú_dataset_trainÚ_dataset_val)r   rƒ   Úsetuprˆ   r   r   Úsetattr)rU   ÚstageÚnameÚdatasetr‰   r‡   s       €€r,   r´   z.train_rfdetr.<locals>.PadAwareDataModule.setup‰   sŽ   ø€ Ý�EÐ0Ð0Ð2`ÑaÔaÐaÝ‰GŒG�MŠM˜%Ñ Ô Ð Ø:ð Nð N�Ý! $¨Ñ-Ô-�ØÐ&­z¸'Å?Ñ/SÔ/SÐ&Ý˜D $­¸ÀÑ(LÔ(LÑMÔMÐMøðNð Nr.   )r¤   r¥   r¦   r´   r§   )r‰   r‡   s   @€r,   ÚPadAwareDataModuler®   ˆ   sH   øø€ € € € € ð	Nð 	Nð 	Nð 	Nð 	Nð 	Nð 	Nð 	Nð 	Nð 	Nr.   r¹   c                  ó&   •— e Zd Zd„ Zd„ Zˆ fd„ZdS )ú*train_rfdetr.<locals>.ResearchEpochReceiptc                óD   — || _         t          j        ¦   «         | _        d S r‹   )ÚattemptÚtimeÚ	monotonicÚstarted)rU   r½   s     r,   r„   z3train_rfdetr.<locals>.ResearchEpochReceipt.__init__’   s   € Ø"ˆDŒLÝœ>Ñ+Ô+ˆDŒLˆLˆLr.   c                ó0  — | j         dz                       ¦   «         syt          | j         dz  t          dt	          |d         ¦  «        |j        t          j                             ¦   «         t          j
        ¦   «         | j        z
  ¬¦  «        ¦  «         d S d S )Nz first_actual_training_batch.jsonÚ0actual_forward_backward_training_batch_succeededr	   )ÚstateÚ
batch_sizeÚglobal_stepÚpeak_cuda_memory_bytesÚseconds_since_start)r½   Úexistsr   ro   rC   rÅ   r   ÚcudaÚmax_memory_allocatedr¾   r¿   rÀ   )rU   ÚtrainerÚ	pl_moduler$   r'   Ú	batch_idxs         r,   Úon_train_batch_endz=train_rfdetr.<locals>.ResearchEpochReceipt.on_train_batch_end–   s¥   € Ø”LÐ!CÑC×KÒKÑMÔMð HÝ�d”lÐ#EÑEÅtØLÝ" 5¨¤8™}œ}¸'Ô:MÝ+0¬:×+JÒ+JÑ+LÔ+LÝ(,¬Ñ(8Ô(8¸¼Ñ(Eð	HGñ HGô HGñ Hô Hð Hð Hð HðHð Hr.   c                ó8  •— d„ |j                              ¦   «         D ¦   «         }|j        dz   }t          | j        dz  |d›d�z  t          ||j        d„ |                     ¦   «         D ¦   «         t          j         	                    ¦   «         t          j        ¦   «         | j        z
  ¬¦  «        ¦  «         t          ‰¦  «         t          t          j        t          d||j        d	„ |                     ¦   «         D ¦   «         ¬
¦  «        ¦  «        d¬¦  «         d S )Nc                ó2  — i | ]”\  }}t          |t          j        ¦  «        r|                     ¦   «         d k    ¯7|t          |t          j        ¦  «        r3t	          |                     ¦   «                              ¦   «         ¦  «        nt	          |¦  «        “Œ•S )r	   )r   r   r   Únumelrm   ÚdetachÚcpu©r;   ÚkÚvs      r,   ú
<dictcomp>zQtrain_rfdetr.<locals>.ResearchEpochReceipt.on_train_epoch_end.<locals>.<dictcomp>Ÿ   s‰   € ð Oð Oð OÙ˜1˜aÝ)¨!­U¬\Ñ:Ô:ðOà>?¿gºg¹i¼iÈ1ºn¸nð µZÀÅ5Ä<Ñ5PÔ5PÐ^�% §¢¡
¤
§¢Ñ 0Ô 0Ñ1Ô1Ð1ÕV[Ð\]ÑV^ÔV^à>L¸n¸nr.   r	   ÚepochsÚ04dz.jsonc                óB   — i | ]\  }}t          j        |¦  «        ¯||“ŒS r9   ©ÚnpÚisfiniterÔ   s      r,   r×   zQtrain_rfdetr.<locals>.ResearchEpochReceipt.on_train_epoch_end.<locals>.<dictcomp>¤   s+   € ÐLÐLÐL¡$ ! Q½R¼[È¹^¼^ÐL˜˜AÐLÐLÐLr.   )ÚepochrÅ   ÚmetricsrÆ   Úelapsed_secondsÚm2c                óB   — i | ]\  }}t          j        |¦  «        ¯||“ŒS r9   rÛ   rÔ   s      r,   r×   zQtrain_rfdetr.<locals>.ResearchEpochReceipt.on_train_epoch_end.<locals>.<dictcomp>©   s,   € Ð*^Ð*^Ð*^±D°A°qÍrÌ{Ð[\É~Ì~Ð*^¨1¨aÐ*^Ð*^Ð*^r.   )ÚrouterÞ   rÅ   rß   T)Úflush)Úcallback_metricsÚitemsÚcurrent_epochr   r½   ro   rÅ   r   rÉ   rÊ   r¾   r¿   rÀ   r   ÚprintÚjsonÚdumps)rU   rË   rÌ   rß   rÞ   Úoutputs        €r,   Úon_train_epoch_endz=train_rfdetr.<locals>.ResearchEpochReceipt.on_train_epoch_endž   sH  ø€ ðOð OØ#*Ô#;×#AÒ#AÑ#CÔ#CðOñ Oô OˆGð Ô)¨!Ñ+ˆEÝ�$”,˜xÑ'¨5Ð(;Ð(;Ð(;Ð(;Ñ;½TÈÐ[bÔ[nØLÐL¨'¯-ª-©/¬/ÐLÑLÔLÝ',¤z×'FÒ'FÑ'HÔ'HÝ $¤Ñ 0Ô 0°´Ñ =ð>?ñ >?ô >?ñ @ô @ð @õ ˜ÑÔÐÝ•$”*�T¨°EÀwÔGZØ*^Ð*^¸G¿MºM¹O¼OÐ*^Ñ*^Ô*^ð`ñ `ô `ñ aô aØhlðnñ nô nð nð nð nr.   N)r¤   r¥   r¦   r„   rÎ   rì   )rë   s   €r,   ÚResearchEpochReceiptr»   ‘   sV   ø€ € € € € ð	,ð 	,ð 	,ð	Hð 	Hð 	Hð	nð 	nð 	nð 	nð 	nð 	nð 	nr.   rí   )r3   r	   é*   T)ÚworkersÚattempt_Ú02diÀ  )rE   rØ   r	   i@  zcuda:0F)Únum_classesÚ
resolutionr4   Úpretrain_weightsÚgradient_checkpointingÚcompileÚdataset_dirÚdataset_rfdetrÚ
output_dirrÄ   Úgrad_accum_stepsé   Úeval_batch_sizeÚlrg-Cëâ6?Ú
lr_encodergñhãˆµøä>ÚseedÚnum_workersÚmulti_scaleÚexpanded_scalesÚscale_jitterÚ
aug_configÚuse_emaÚearly_stoppingÚearly_stopping_patienceÚearly_stopping_min_deltagü©ñÒMbP?Úearly_stopping_use_emaÚrun_testÚtensorboardÚwandbÚprogress_barÚcheckpoint_intervaliè  Úcompute_val_lossr‚   Ú	fork_faceÚsquare_resize_div_64Úeval_masks_head_resolutionÚacceleratorÚgpuÚdevicesÚ
pin_memoryÚ	amp_dtypeÚautoÚsave_dataset_gridszrun_config.jsonré   zNOfficial RFDETRSegSmall / RFDETRModelModule / RFDETRDataModule / build_trainerz„preletterboxed nearest RLE; padding ignored in mask matcher and CE/Dice; dense mask-head loss replaces upstream uncertainty sampling)ÚmodelÚtrainingÚgeneric_provenanceÚapiÚpadding_adapterÚexpected_effective_batchÚoptimizer_archive_interval)r  r  Ú	completed)rÄ   ÚaccumulationrÃ   rÆ   ÚsecondsÚactual_cuda_oomzoom.jsonr@   zmemory_profile.json)ÚattemptsÚselected_batchÚrestart_from_generic_on_oomÚfailed_attempts_retainedz1M2 OOM at batch 1; no training completion claimedzcheckpoint_best_total.pthz:Official RF-DETR trainer did not produce a best checkpoint)rò   ró   r4   rô   rö   Údevelopment_splitÚdevr_   r˜   z Missing internal development ROIg        )Ú	thresholdÚinclude_source_imageÚnative_maskg      à?zfrozen_selection.jsonzVupstream best EMA internal grouped development mask AP50:95, patience15 min_delta .001Ú quality_filter_confidence_cutoff)Ú
checkpointÚsha256Ú	selectionÚnative_mask_thresholdÚconfidence_cutoffÚofficial_val_test_evaluatedÚ2model_frozen_before_any_future_official_evaluationÚall)Úepochs_trainedr/  Úcheckpoint_sha256Úinternal_qualityÚphysical_batchr9   )EÚpytorch_lightningÚrfdetrr©   Úrfdetr.trainingrª   r«   r¬   r   ÚCallbackÚ	enumerater   Úseed_everythingÚmkdirr’   Úget_train_configr   r   Úmodel_configr   ro   Ú
model_dumpÚgcÚcollectr   rÉ   Úempty_cacheÚtypesÚ
MethodTyper}   Ú	criterionÚ
loss_masksre   ÚmatcherÚ_compute_mask_costsÚ	callbacksr"   Úreset_peak_memory_statsr¾   r¿   r°   rÊ   rç   ÚOutOfMemoryErrorr   Úis_filer   Úcv2Úimreadr   r·   ÚsynchronizeÚpredictÚIMREAD_GRAYSCALErC   Úmaskr“   rÜ   ÚargmaxÚ
confidencer   ÚastypeÚfloat32rm   Ú
zeros_liker   r   r   Úrelative_tor
   )'rë   ÚoutÚrecordsÚweightsÚ
provenanceÚplr©   rª   r«   r¬   r¹   rí   r%  Úcompleted_attemptr*   r'   r½   r  Útrain_configrC  ÚmoduleÚdatarË   rÀ   r7  ÚbestÚprediction_recordsÚrecordr˜   ÚstartÚ
detectionsÚdurationrv   r+   rc   rW  rY  Úreportr‡   s'   `                                     @r,   Útrain_rfdetrrn  �   s­	  øø€ Ø"Ð"Ð"Ð"Ø%Ð%Ð%Ð%Ð%Ð%ØRÐRÐRÐRÐRÐRÐRÐRÐRÐRå�fÐ=Ñ=Ñ>Ô>€JðNð Nð Nð Nð Nð Nð NÐ-ñ Nô Nð Nðnð nð nð nð nð nð n˜rœ{ñ nô nð nð4 €HØÐÝ! &Ñ)Ô)ð 83ñ 83‰ˆˆuÝ‰
Œ
ˆ
Ø
×Ò˜2 tÐÑ,Ô,Ð,ØÐ, Ð,Ð,Ð,Ñ,ˆØ�Š˜5ˆÑ!Ô!Ð!Ø	�Ñ	× Ò  eÐ Ñ,Ô,Ð,Ø�¨1¸ÀXÝ  ™\œ\À$ÐPUðWñ Wô Wˆà-�uÔ-ð Ið Ið IÝ˜FÐ#3Ñ3Ñ4Ô4Ð4ðIÝADÀWÁÄÀðIå�:ðIà*/¨%ðIàBCÀUÁ(À(ðIð ˜AðIð #' $ðIð 48°4ðIð ?A¸bðIð ˜ð	Ið (- uð	Ið ?D¸eð	Ið
 ˜ðIð
 ,.¨2ðIð
 8<°tðIð
 MQÈDðIõ %- HðIð HLÀtðIð $( 4ðIð 38°%ðIð FKÀUðIð �%ðIð '+ dðIð AEÀðIð "˜TðIð 1<¨}¨}ðIð TXÐSWðIð (- uðIð ;@¸%ðIð JKÈðIð �tðIð (. vðIð CHÀ%ðIˆð Ô)ˆÝˆgÐ'Ñ'­°L×4KÒ4KÐQWÐ4KÑ4XÔ4XØ!×,Ò,°&Ð,Ñ9Ô9ÈjØ`ðZà%&À4ð*Iñ *Iô *Iñ 	Jô 	Jð 	Jð Ý
Œ
‰Œˆ•e”j×,Ò,Ñ.Ô.Ð.Ø"Ð" <°Ñ>Ô>ˆÝ&+Ô&6Õ7NÐPVÔP`Ñ&aÔ&aˆÔÔ#Ý7<Ô7GÝ$ fÔ&6Ô&>ñ8@ô 8@ˆÔÔ Ô4à!Ð! ,°Ñ=Ô=ˆð  �- ¨lÈÐWXÐYÑYÔYˆØÔ× Ò Ð!5Ð!5°gÑ!>Ô!>Ñ?Ô?Ð?ÝŒ
×*Ò*Ñ,Ô,Ð,Ý”.Ñ"Ô"ˆð	3Ø�KŠK˜ Ñ%Ô%Ð%Ø�OŠO�D¨EÀÀ5ÁÐP[Ý',¤z×'FÒ'FÑ'HÔ'HÕRVÔR`ÑRbÔRbÐcjÑRjðlñ lô lñ mô mð mà 'ÐØ$Ô2ˆNØ˜˜gÝŒJ‰LŒLˆL�%œ*×0Ò0Ñ2Ô2Ð2ØˆEøÝŒzÔ*ð 	3ð 	3ð 	3Ø�OŠO�D¨EÀÀ5ÁÐPaÝ',¤z×'FÒ'FÑ'HÔ'HÕRVÔR`ÑRbÔRbÐcjÑRjðlñ lô lñ mô mð må�'˜*Ñ$ h¨r¤lÑ3Ô3Ð3Ø˜˜gÝŒJ‰LŒLˆL�%œ*×0Ò0Ñ2Ô2Ð2Ð2Ñ2ð	3øøøõ 
ˆ#Ð#Ñ
#¥T°8Ø5FÐP�x ”| LÔ1Ð1ÈDØ$(À4ð&Iñ &Iô &Iñ Jô Jð Jõ Ð TÐ)Ð+^Ñ_Ô_Ð_ØÐ8Ñ8€DÝˆD�LŠL‰NŒNÐXÑYÔYÐYÝ�6ÑÔÐØˆN q°SÀÝ+.¨t©9¬9¸eðEñ Eô E€EàÐÝ" 7Ñ+Ô+ð mñ m‰ˆˆvØÐ%Ô&¨%Ò/Ð/ØÝ”
�3˜vÐ&6Ñ6¸Ñ@ÅÀfÈWÄoÑAVÔAVÔA[Ñ[Ñ\Ô\Ñ]Ô]ˆÝ�˜TÐ!Ð#EÑFÔFÐFÝŒ
×ÒÑ Ô Ð Ý”Ñ Ô ˆØ—]’] 5°BÈU�]ÑSÔSˆ
ÝŒ
×ÒÑ Ô Ð Ý”>Ñ#Ô# EÑ)ˆÝ”�C  v¨mÔ'<Ñ <Ñ=Ô=½sÔ?SÑTÔTÐWXÒXˆÝˆz‰?Œ?ð 	9˜zœÐ:å�œ :Ô#8Ñ9Ô9Ñ:Ô:ˆFÝ% j¤o°fÔ&=×&DÒ&DÅRÄZÑ&PÔ&PÐR\Õ]`ÐafÐghÑahÑ]iÔ]iÔRjÑkÔkˆKØ "Ò$ˆDÝ˜zÔ4°VÔ<Ñ=Ô=ˆJˆJå!œ}¨VÑ4Ô4°b�*ˆDØ×!Ò!¥-°½ÀtÈVÑ8TÔ8TÐV`ÐbjÑ"kÔ"kÑlÔlÐlÑlÝ˜CÐ!3Ñ4Ô4€FÝ	ˆ#Ð%Ñ
%¥tµs¸4×;KÒ;KÈCÑ;PÔ;PÑ7QÔ7QÕZ`ÐaeÑZfÔZfØjØ °FÐ;]Ô4^Ø$)Ð^bð(dñ (dô (dñ eô eð eõ ˜~½#¸d×>NÒ>NÈsÑ>SÔ>SÑ:TÔ:TÝ"(¨¡,¤,ÀÈÄØ'¨œ|¨LÔ9ð;ñ ;ô ;ð ;s   ËB-N
Î
B;Q	ÑQ	r‹   ))Ú__doc__Ú
__future__r   rE  ré   Úpathlibr   r¾   rH  rR  ÚnumpyrÜ   r   r   Útorch.nnr   rO   Útorch.utils.datar   r   Úmask_model_training_datar
   r   r   r   r   r   r   Úmask_model_trainingr   r   r   r   r   r   r-   re   rr   r}   r   rn  r9   r.   r,   ú<module>rw     sð  ððð ð #Ð "Ð "Ð "Ð "Ð "à 	€	€	€	Ø €€€Ø Ð Ð Ð Ð Ð Ø €€€Ø €€€à 
€
€
€
Ø Ð Ð Ð Ø €€€Ø Ð Ð Ð Ð Ð Ø $Ð $Ð $Ð $Ð $Ð $Ø ,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 ð ð ð ð ð ð ð ð ð ð ð ð ð ð
Kð Kð Kð Kð ð ð ð4Vð Vð Vð	Að 	Að 	Að2ð 2ð 2ð 2ð 2�fñ 2ô 2ð 2ð:J;ð J;ð J;ð J;ð J;r.   