§
    ­¸j41  ã                   ó–  — d Z ddlZddlZddlZddlmZ ddlmZm	Z	m
Z
 ddlmZmZmZ ddlZddlmZ ddlmZ ddlmZ 	 ddlZn# e$ r dZY nw xY w	 ddlmZ n# e$ r dZY nw xY wddlmZmZ dd	lmZ  e¦   «         Z  G d
„ de¦  «        Z!de"de#e$e"ef                  fd„Z%de"dej        fd„Z&dZ'dZ(dede)de)fd„Z*d8dede$e"ef         dz  de+ee$e"ef         dz  f         fd„Z,e'fde+e)e)f         defd„Z-dedefd„Z.e(fde	e"ef         dede)de#e$e"ef                  fd „Z/	 d9d"ed#ed$d%d&e"d'e
e$e"ef                  d(e"ej0        z  d)e)ddfd*„Z1	 d9d+d,d#ed$d%d&e"d'e
e$e"ef                  d(e"ej0        z  d)e)ddfd-„Z2	 	 	 	 d:d0e"d1e"d(e)d2e3d3e3ddfd4„Z4e5d5k    r1 ej6        d6¦  «        Z7  ee! e8e7d7¦  «        ¦  «        e4¦  «         dS dS );zñThis tool provides performance benchmarks by using ONNX Runtime and TensorRT to run inference on a given model with
the COCO validation set.

It offers reliable measurements of inference latency using ONNX Runtime or TensorRT on the device.
é    N)ÚCallableÚMappingÚSequence)ÚAnyÚProtocolÚcast)ÚImage)ÚTensor)Útqdm)ÚTimeProfilerÚTRTInference)Ú
get_loggerc                   ó2   — e Zd ZdZdedef         defd„ZdS )Ú_JsonArgparseCLIz@Minimal jsonargparse CLI callable interface used by this script.Ú	component.Úreturnc                 ó   — dS )z.Run jsonargparse CLI for a callable component.N© )Úselfr   s     úg/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/rfdetr/export/benchmark.pyÚ__call__z_JsonArgparseCLI.__call__.   s   € àˆó    N)Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   r   r   r   r   r   r   r   +   sE   € € € € € ØJÐJð (¨3°¨8Ô"4ð ¸ð ð ð ð ð ð r   r   Úann_filer   c                 ó¢   — t          | ¦  «        5 }t          j        |¦  «        }d d d ¦  «         n# 1 swxY w Y   t          |d         ¦  «        S )NÚimages)ÚopenÚjsonÚloadÚlist)r   ÚfinÚdatas      r   Úget_image_listr&   3   s   € Ý	ˆh‰Œð ˜3ÝŒy˜‰~Œ~ˆðð ð ñ ô ð ð ð ð ð ð øøøð ð ð ð å��X”ÑÔÐs   �1±5¸5Ú	file_pathc                 óP   — t          j        | ¦  «                             d¦  «        S )NÚRGB)r	   r    Úconvert)r'   s    r   Ú
load_imager+   9   s    € ÝŒ:�iÑ Ô ×(Ò(¨Ñ/Ô/Ð/r   )é€  r,   i,  ÚvalueÚfallbackc                 ój   — 	 t          | ¦  «        }n# t          t          f$ r |cY S w xY w|dk    r|n|S )a  Coerce a tensor-shape entry to a positive int, falling back for dynamic/unknown axes.

    ONNX Runtime and TensorRT report dynamic axes as strings (e.g. ``"height"``), ``None``, or ``-1``. Such values
    cannot drive a fixed preprocessing size, so the caller-supplied *fallback* is used instead.

    Args:
        value: A single entry from a model input/output shape.
        fallback: Value to return when *value* is not a concrete positive integer.

    Returns:
        The integer dimension, or *fallback* for dynamic axes.
    r   )ÚintÚ	TypeErrorÚ
ValueError)r-   r.   Údims      r   Ú_static_dimr4   A   sO   € ðÝ�%‰jŒjˆˆøÝ•zÐ"ð ð ð Øˆˆˆðøøøà˜’'�'ˆ3ˆ3˜xÐ's   ‚ ’(§(ÚimageÚtargetc                 ó.   — |                       ¦   «         |fS )u¡  Materialize *image* as a contiguous tensor, leaving *target* untouched.

    ``ToImage`` permutes a decoded HWC buffer to CHW without copying, so the pipeline otherwise
    yields a channels_last view. Runtimes that read the input buffer directly rather than honoring
    strides â€” the ExecuTorch runtime among them â€” then misread the image and silently return wrong
    predictions, so the copy is forced here where every export inference path picks it up.

    Args:
        image: CHW image tensor, possibly a non-contiguous view.
        target: Optional annotation dict, passed through unchanged.

    Returns:
        Tuple of ``(contiguous_image, target)``.
    )Ú
contiguous)r5   r6   s     r   Ú_ensure_contiguousr9   U   s   € ð ×ÒÑÔ˜vÐ%Ð%r   Úsizec                 ó°   — ddl m}m}m}m} ddlm}  | |¦   «          |t          j        d¬¦  «         || d¬¦  «         |¦   «         t          g¦  «        S )aÈ  Build the benchmark preprocessing pipeline for a given model input size.

    Args:
        size: Target ``(height, width)`` the image is resized to before inference. Defaults to
            :data:`_DEFAULT_INPUT_SIZE` for dynamic-axis models where a static size cannot be read.

    Returns:
        A ``torchvision.transforms.v2.Compose`` that tensorizes, resizes, normalizes, and returns
        the image as a contiguous tensor.

    Note:
        Tensorize-then-resize with ``antialias=False`` mirrors ``RFDETR.predict()``'s preprocessing
        (``detr.py``): resizing the PIL image first would apply PIL's adaptive antialias filter and
        benchmark the model on inputs predict() never produces.
    r   )ÚComposeÚResizeÚToDtypeÚToImage)Ú	NormalizeT)ÚscaleF)Ú	antialias)
Útorchvision.transforms.v2r<   r=   r>   r?   Úrfdetr.datasets.transformsr@   ÚtorchÚfloat32r9   )r:   r<   r=   r>   r?   r@   s         r   Úinfer_transformsrG   g   s�   € ð  LÐKÐKÐKÐKÐKÐKÐKÐKÐKÐKÐKà4Ð4Ð4Ð4Ð4Ð4àˆ7àˆG‰IŒIØˆG•E”M¨Ð.Ñ.Ô.ØˆF�4 5Ð)Ñ)Ô)ØˆI‰KŒKÝð	
ñô ð r   Úxc                 ó>  — |                       d¦  «        \  }}}}|d|                     d¬¦  «        z  z
  |d|                     d¬¦  «        z  z
  |d|                     d¬¦  «        z  z   |d|                     d¬¦  «        z  z   g}t          j        |d¬¦  «        S )Néÿÿÿÿg      à?g        )Úmin©r3   )ÚunbindÚclamprE   Ústack)rH   Úx_cÚy_cÚwÚhÚbs         r   Úbox_cxcywh_to_xyxyrU   †   s£   € Ø—X’X˜b‘\”\�N€Cˆˆa�à	ˆs�Q—W’W �WÑ%Ô%Ñ%Ñ	%Ø	ˆs�Q—W’W �WÑ%Ô%Ñ%Ñ	%Ø	ˆs�Q—W’W �WÑ%Ô%Ñ%Ñ	%Ø	ˆs�Q—W’W �WÑ%Ô%Ñ%Ñ	%ð		€Aõ Œ;�q˜bÐ!Ñ!Ô!Ð!r   ÚoutputsÚtarget_sizesÚnum_queriesc           	      ó   — | d         | d         }}t          |¦  «        t          |¦  «        k    sJ ‚|j        d         dk    sJ ‚|                     ¦   «         }|                     |j        d         d¦  «        }t	          ||j        d         ¦  «        }t          j        ||d¬¦  «        \  }}	|}
|	|j        d         z  }|	|j        d         z  }t          |¦  «        }t          j        |d| 	                    d¦  «         
                    ddd¦  «        ¦  «        }|                     d¦  «        \  }}t          j        ||||gd¬¦  «        }||d d …d d d …f         z  }d	„ t          |
||¦  «        D ¦   «         }|S )
NÚlabelsÚdetsé   é   r   rJ   rL   é   c                 ó"   — g | ]\  }}}|||d œ‘ŒS ))ÚscoresrZ   Úboxesr   )Ú.0ÚscoreÚlabelÚboxs       r   ú
<listcomp>z post_process.<locals>.<listcomp>ª   s*   € ÐtÐtÐtÑDUÀEÈ5ÐRU˜%¨5¸3Ð?Ð?ÐtÐtÐtr   )ÚlenÚshapeÚsigmoidÚviewrK   rE   ÚtopkrU   ÚgatherÚ	unsqueezeÚexpandrM   rO   Úzip)rV   rW   rX   Ú
out_logitsÚout_bboxÚprobÚflat_scoresÚkÚtopk_valuesÚtopk_indexesr`   Ú
topk_boxesrZ   ra   Úimg_hÚimg_wÚ	scale_fctÚresultss                     r   Úpost_processr|   ‘   s’  € ð # 8Ô,¨g°f¬o�€Jåˆz‰?Œ?�c ,Ñ/Ô/Ò/Ð/Ð/Ð/ØÔ˜aÔ  AÒ%Ð%Ð%Ð%à×ÒÑÔ€DØ—)’)˜JÔ,¨QÔ/°Ñ4Ô4€Kõ 	ˆK˜Ô*¨1Ô-Ñ.Ô.€AÝ %¤
¨;¸¸qÐ AÑ AÔ AÑ€K�Ø€FØ Ô!1°!Ô!4Ñ4€JØ˜JÔ,¨QÔ/Ñ/€FÝ˜xÑ(Ô(€EÝŒL˜  :×#7Ò#7¸Ñ#;Ô#;×#BÒ#BÀ2ÀrÈ1Ñ#MÔ#MÑNÔN€Eð  ×&Ò& qÑ)Ô)�L€Eˆ5Ý”˜U E¨5°%Ð8¸aÐ@Ñ@Ô@€IØ�I˜a˜a˜a  q q q˜jÔ)Ñ)€EàtÐtÕY\Ð]cÐekÐmrÑYsÔYsÐtÑtÔt€Gà€Nr   r\   ÚsessÚcoco_evaluatorÚtime_profiler   ÚprefixÚimg_listÚdeviceÚrepeatsc                 óø  — |                       ¦   «         d         j        }t          t          |¦  «        dk    r|d         nd t          d         ¦  «        }t          t          |¦  «        dk    r|d         nd t          d         ¦  «        }	|                      ¦   «         d         j        }
t          t          |
¦  «        dk    r|
d         nd t          ¦  «        }t          ||	f¦  «        }g }t          |¦  «        D �]º}t          t          j                             ||d         ¦  «        ¦  «        }|j        \  }}t          j        ||g¦  «        } ||d ¦  «        \  }}|d                               ¦   «         }|                     ¦   «          |5  t%          |¦  «        D ]}|                      d d|i¦  «        }Œ	 d d d ¦  «         n# 1 swxY w Y   |                     |j        |z  ¦  «         i }t          j        |d         ¦  «                             |¦  «        |d<   t          j        |d         ¦  «                             |¦  «        |d<   t          j        |gd¬	¦  «                             |¦  «        }t1          |||¬
¦  «        }|d         |d         i}|�|                     |¦  «         �Œ¼t4                               ddt9          |¦  «        z  t          |¦  «        z  › d�¦  «         i }|�|                     ¦   «          |                     ¦   «          |                     ¦   «          |j         d         j!         "                    ¦   «         |d<   t4                               |¦  «         d S d S )Nr   é   r]   r\   Ú	file_nameÚinputrZ   r[   rL   ©rX   Úidz!Model latency with ONNX Runtime: éè  ÚmsÚbboxÚcoco_eval_bbox)#Ú
get_inputsrh   r4   rg   Ú_DEFAULT_INPUT_SIZEÚget_outputsÚ_DEFAULT_NUM_QUERIESrG   r   r+   ÚosÚpathÚjoinr:   rE   r
   ÚnumpyÚresetÚrangeÚrunÚappendÚtotalÚtorO   r|   ÚupdateÚloggerÚinfoÚsumÚsynchronize_between_processesÚ
accumulateÚ	summarizeÚ	coco_evalÚstatsÚtolist)r}   r~   r   r€   r�   r‚   rƒ   Úinput_shapeÚinput_hÚinput_wÚoutput_shaperX   Ú
transformsÚ	time_listÚimg_dictr5   ÚwidthÚheightÚorig_target_sizesÚimage_tensorÚ_ÚsamplesÚresrV   r{   r¤   s                             r   Ú
infer_onnxr´   ¯   sˆ  € ð —/’/Ñ#Ô# AÔ&Ô,€Kå­C°Ñ,<Ô,<¸qÒ,@Ð,@˜+ aœ.˜.ÀdÕL_Ð`aÔLbÑcÔc€GÝ­C°Ñ,<Ô,<¸qÒ,@Ð,@˜+ aœ.˜.ÀdÕL_Ð`aÔLbÑcÔc€GØ×#Ò#Ñ%Ô% aÔ(Ô.€LÝµ°\Ñ1BÔ1BÀQÒ1FÐ1F˜l¨1œo˜oÈDÕRfÑgÔg€KÝ! 7¨GÐ"4Ñ5Ô5€Jà€IÝ˜‘N”Nð 'ñ 'ˆÝ�2œ7Ÿ<š<¨°¸Ô0EÑFÔFÑGÔGˆØœ
‰ˆˆvÝ!œL¨&°%¨Ñ9Ô9ÐØ$˜* U¨DÑ1Ô1‰ˆ�aà˜tÔ$×*Ò*Ñ,Ô,ˆà×ÒÑÔÐØð 	9ð 	9Ý˜7‘^”^ð 9ð 9�Ø—h’h˜t g¨wÐ%7Ñ8Ô8��ð9ð	9ð 	9ð 	9ñ 	9ô 	9ð 	9ð 	9ð 	9ð 	9ð 	9ð 	9øøøð 	9ð 	9ð 	9ð 	9ð 	×Ò˜Ô+¨gÑ5Ñ6Ô6Ð6ØˆÝ!œL¨¨Q¬Ñ0Ô0×3Ò3°FÑ;Ô;ˆ�ÑÝœ, s¨1¤vÑ.Ô.×1Ò1°&Ñ9Ô9ˆ�‰å!œKÐ):Ð(;ÀÐCÑCÔC×FÒFÀvÑNÔNÐÝ˜wÐ(9À{ÐSÑSÔSˆØ˜Œ~˜w qœzÐ*ˆØÐ%Ø×!Ò! #Ñ&Ô&Ð&ùå
‡K‚KÐ]°D½3¸y¹>¼>Ñ4IÍCÐPXÉMÌMÑ4YÐ]Ð]Ð]Ñ^Ô^Ð^ð €EØÐ!Ø×4Ò4Ñ6Ô6Ð6Ø×!Ò!Ñ#Ô#Ð#Ø× Ò Ñ"Ô"Ð"Ø"0Ô":¸6Ô"BÔ"H×"OÒ"OÑ"QÔ"QˆÐÑÝ�Š�EÑÔÐÐÐð "Ð!s   Æ+G
Ç
G	ÇG	Úmodelr   c                 ód  — t          | j        | j        d                  j        ¦  «        }t	          t          |¦  «        dk    r|d         nd t          d         ¦  «        }t	          t          |¦  «        dk    r|d         nd t          d         ¦  «        }	t          | j        | j        d                  j        ¦  «        }
t	          t          |
¦  «        dk    r|
d         nd t          ¦  «        }t          ||	f¦  «        }g }t          |¦  «        D �]Z}t          t          j                             ||d         ¦  «        ¦  «        }|j        \  }}t!          j        ||g¦  «        } ||d ¦  «        \  }}|d                               |¦  «        }|j        \  }}}}|                     ¦   «          |5  t)          |¦  «        D ]} | d|i¦  «        }Œ	 d d d ¦  «         n# 1 swxY w Y   |                     |j        |z  ¦  «         t!          j        |gd¬¦  «                             |¦  «        }|�7t1          |||¬¦  «        }|d	         |d         i}|                     |¦  «         �Œ\t4                               d
dt9          |¦  «        z  t          |¦  «        z  › d�¦  «         i }|�|                     ¦   «          |                     ¦   «          |                     ¦   «          |j         d         j!         "                    ¦   «         |d<   t4                               |¦  «         d S d S )Nr   r…   r]   r\   r†   r‡   rL   rˆ   r‰   zModel latency with TensorRT: rŠ   r‹   rŒ   r�   )#r#   ÚbindingsÚinput_namesrh   r4   rg   r�   Úoutput_namesr‘   rG   r   r+   r’   r“   r”   r:   rE   r
   r›   r–   r—   r™   rš   rO   r|   rœ   r�   rž   rŸ   r    r¡   r¢   r£   r¤   r¥   )rµ   r~   r   r€   r�   r‚   rƒ   r¦   r§   r¨   r©   rX   rª   r«   r¬   r5   r­   r®   r¯   r°   r±   r²   rS   rR   rV   r{   r³   r¤   s                               r   Úinfer_enginerº   ä   sT  € õ �u”~ eÔ&7¸Ô&:Ô;ÔAÑBÔB€Kå­C°Ñ,<Ô,<¸qÒ,@Ð,@˜+ aœ.˜.ÀdÕL_Ð`aÔLbÑcÔc€GÝ­C°Ñ,<Ô,<¸qÒ,@Ð,@˜+ aœ.˜.ÀdÕL_Ð`aÔLbÑcÔc€GÝ˜œ uÔ'9¸!Ô'<Ô=ÔCÑDÔD€LÝµ°\Ñ1BÔ1BÀQÒ1FÐ1F˜l¨1œo˜oÈDÕRfÑgÔg€KÝ! 7¨GÐ"4Ñ5Ô5€Jà€IÝ˜‘N”Nð 'ñ 'ˆÝ�2œ7Ÿ<š<¨°¸Ô0EÑFÔFÑGÔGˆØœ
‰ˆˆvÝ!œL¨&°%¨Ñ9Ô9ÐØ$˜* U¨DÑ1Ô1‰ˆ�aà˜tÔ$×'Ò'¨Ñ/Ô/ˆØ”]‰
ˆˆ1ˆa�ð 	×ÒÑÔÐØð 	4ð 	4Ý˜7‘^”^ð 4ð 4�Ø˜% ¨'Ð 2Ñ3Ô3��ð4ð	4ð 	4ð 	4ñ 	4ô 	4ð 	4ð 	4ð 	4ð 	4ð 	4ð 	4øøøð 	4ð 	4ð 	4ð 	4ð 	×Ò˜Ô+¨gÑ5Ñ6Ô6Ð6Ý!œKÐ):Ð(;ÀÐCÑCÔC×FÒFÀvÑNÔNÐØÐ%Ý" 7Ð,=È;ÐWÑWÔWˆGØ˜D”> 7¨1¤:Ð.ˆCØ×!Ò! #Ñ&Ô&Ð&ùå
‡K‚KÐY°µs¸9±~´~Ñ0EÍÈHÉÌÑ0UÐYÐYÐYÑZÔZÐZð €EØÐ!Ø×4Ò4Ñ6Ô6Ð6Ø×!Ò!Ñ#Ô#Ð#Ø× Ò Ñ"Ô"Ð"Ø"0Ô":¸6Ô"BÔ"H×"OÒ"OÑ"QÔ"QˆÐÑÝ�Š�EÑÔÐÐÐð "Ð!s   Æ5 G"Ç"G&	Ç)G&	ú	data/cocoFr“   Ú	coco_pathÚrun_benchmarkÚdisable_evalc           	      ó¨  — t                                | ||||dœ¦  «         t          j        |d¦  «        }t	          |¦  «        }t          j        |d¦  «        }|rd}t                                d¦  «         nd}|s#ddlm}	 dd	lm}
  |
 |	|¦  «        d
g¦  «        }nd}t          ¦   «         }|  
                    d¦  «        r;ddl}|                     | dd|ifg¬¦  «        }t          |||||d|› �|¬¦  «         dS |  
                    d¦  «        r0t          | dd|› �¬¦  «        }t          |||||d|› �|¬¦  «         dS t!          d¦  «        ‚)a,  Performance benchmark tool for ONNX/TRT models.

    Args:
        path: Engine file path (.onnx or .trt/.engine).
        coco_path: COCO dataset path.
        device: CUDA device index.
        run_benchmark: Repeat inference 10x to measure latency.
        disable_eval: Skip COCO evaluation.
    )r“   r¼   r‚   r½   r¾   z"annotations/instances_val2017.jsonÚval2017é
   z6Inference for each image will be repeated 10 times ...r\   r   )ÚCOCO)ÚCocoEvaluatorrŒ   Nz.onnxÚCUDAExecutionProviderÚ	device_id)Ú	providerszcuda:)r‚   rƒ   )z.trtz.engineT)Ú	sync_moder‚   zNOnly model file names ending with ".onnx", ".trt", or ".engine" are supported.)r�   rž   Úospr”   r&   Úfaster_coco_evalrÂ   Úrfdetr.evaluation.coco_evalrÃ   r   ÚendswithÚonnxruntimeÚInferenceSessionr´   r   rº   ÚNotImplementedError)r“   r¼   r‚   r½   r¾   Úcoco_gtr�   r€   rƒ   rÂ   rÃ   r~   r   Únxrunr}   rµ   s                   r   ÚmainrÑ     sÒ  € õ  ‡K‚KàØ"ØØ*Ø(ð	
ð 	
ñô ð õ Œh�yÐ"FÑGÔG€GÝ˜gÑ&Ô&€HÝŒX�i Ñ+Ô+€FØð ØˆÝ�ŠÐLÑMÔMÐMÐMàˆàð Ø)Ð)Ð)Ð)Ð)Ð)à=Ð=Ð=Ð=Ð=Ð=à&˜ t t¨G¡}¤}°v°hÑ?Ô?ˆˆàˆÝ‘>”>€Là‡}‚}�WÑÔð tØ#Ð#Ð#Ð#à×%Ò%ØØ/°+¸vÐ1FÐGÐHð &ñ 
ô 
ˆõ 	�4˜¨°v¸xÐP`ÐX^ÐP`ÐP`ÐjqÐrÑrÔrÐrÐrÐrØ	�ŠÐ*Ñ	+Ô	+ð tÝ˜T¨TÐ:JÀ&Ð:JÐ:JÐKÑKÔKˆÝ�U˜N¨L¸&À(ÐScÐ[aÐScÐScÐmtÐuÑuÔuÐuÐuÐuå!Ð"rÑsÔsÐsr   Ú__main__ÚjsonargparseÚCLI)N)r\   )r»   r   FF)9r   Ú	importlibr!   r’   Úos.pathr“   rÈ   Úcollections.abcr   r   r   Útypingr   r   r   rE   ÚPILr	   r
   Ú	tqdm.autor   ÚtensorrtÚtrtÚImportErrorÚpycuda.driverÚdriverÚcudaÚ!rfdetr.export._tensorrt.inferencer   r   Úrfdetr.utilities.loggerr   r�   r   Ústrr#   Údictr&   r+   r�   r‘   r0   r4   Útupler9   rG   rU   r|   r‚   r´   rº   ÚboolrÑ   r   Úimport_modulerÓ   Úgetattrr   r   r   ú<module>ré      så  ððð ð Ð Ð Ð Ø €€€Ø 	€	€	€	Ø Ð Ð Ð Ð Ð Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø &Ð &Ð &Ð &Ð &Ð &Ð &Ð &Ð &Ð &à €€€Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð ðØÐÐÐÐøØð ð ð Ø
€C€C€CðøøøðØ Ð Ð Ð Ð Ð Ð øØð ð ð Ø€D€D€Dðøøøð IÐ HÐ HÐ HÐ HÐ HÐ HÐ HØ .Ð .Ð .Ð .Ð .Ð .à	ˆ‰Œ€ðð ð ð ð �xñ ô ð ð ˜Sð   T¨$¨s°C¨x¬.Ô%9ð  ð  ð  ð  ð0˜#ð 0 %¤+ð 0ð 0ð 0ð 0ð !Ð ØÐ ð(�sð ( cð (¨cð (ð (ð (ð (ð(&ð &˜fð &¨d°3¸°8¬n¸tÑ.Cð &ÈuÐU[Ð]aÐbeÐgjÐbjÔ]kÐnrÑ]rÐUrÔOsð &ð &ð &ð &ð$ .Að ð ˜5  c œ?ð ÀSð ð ð ð ð>"˜&ð " Vð "ð "ð "ð "ð Maðð Ø�S˜&�[Ô!ðØ17ðØFIðà	ˆ$ˆs�Fˆ{Ô
Ôðð ð ð ðJ ð2ð 2Ø
ð2àð2ð !ð2ð ð	2ð
 �t˜C ˜H”~Ô&ð2ð �%”,Ñð2ð ð2ð 
ð2ð 2ð 2ð 2ðx ð2ð 2Øð2àð2ð !ð2ð ð	2ð
 �t˜C ˜H”~Ô&ð2ð �%”,Ñð2ð ð2ð 
ð2ð 2ð 2ð 2ðn !ØØØð8tð 8tØ
ð8tàð8tð ð8tð ð	8tð
 ð8tð 
ð8tð 8tð 8tð 8tðv ˆzÒÐØ*�9Ô*¨>Ñ:Ô:€Là8€D€DÐ	˜7˜7 <°Ñ7Ô7Ñ8Ô8¸Ñ>Ô>Ð>Ð>Ð>ð Ðs$   Á A ÁAÁAÁA ÁA$Á#A$