§
    l¬¸jƒé  ã                  ó.  — d dl mZ d dlZd dlmZ d dlmZ d dlmZ d dl	m
Z
 d dlm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mZmZmZ d dlmZ d dlmZmZ d dl m!Z! ddl"m#Z#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z)m*Z*m+Z+ ddl,m-Z- ddl.m/Z/ ddl0m1Z1m2Z2m3Z3m4Z4m5Z5m6Z6m7Z7m8Z8m9Z9m:Z:m;Z;m<Z<m=Z= dZ> G d„ de-¦  «        Z? G d„ de?¦  «        Z@ G d„ de?¦  «        ZA G d„ de?¦  «        ZB G d„ de¦  «        ZC G d„ de?¦  «        ZD G d „ d!eDe?¦  «        ZE G d"„ d#¦  «        ZFdS )$é    )ÚannotationsN)Údefaultdict)Úrepeat)Ú
ThreadPool)ÚPath)ÚAny)ÚImage)ÚConcatDataset)Ú
LOCAL_RANKÚLOGGERÚNUM_THREADSÚTQDMÚcolorstr)Ú	Instances)Úresample_segmentsÚsegments2boxes)ÚTORCHVISION_0_18é   )	ÚComposeÚDepthFormatÚFormatÚ	LetterBoxÚRandomLoadTextÚSemanticFormatÚclassify_augmentationsÚclassify_transformsÚv8_transforms)ÚBaseDataset)Úmerge_multi_segment)ÚHELP_URLÚcheck_file_speedsÚget_hashÚget_split_fractionÚimg2label_pathsÚload_dataset_cache_fileÚ
load_depthÚpolygons2masks_overlapÚsave_dataset_cache_fileÚverify_imageÚverify_image_depthÚverify_image_labelÚverify_image_maskz1.0.4c                  óà   ‡ — e Zd ZdZeZdddœd5ˆ fd	„Z ed
¦  «        fd6d„Zd7d„Z	d8d„Z
d9d„Zd:d„Zd;d„Zd<d#„Zd=d&„Zd>d'„Zd?d@d*„ZdAd-„ZedBd/„¦   «         ZdCd0„ZdDd2„ZedEd4„¦   «         Zˆ xZS )FÚYOLODatasetaI  Dataset class for loading object detection and/or segmentation labels in YOLO format.

    This class supports loading data for object detection, instance segmentation, pose estimation, and oriented bounding
    box (OBB) tasks using the YOLO format.

    Attributes:
        format_class (type[Format]): Formatter appended by build_transforms; subclasses override it per task.
        use_segments (bool): Indicates if segmentation masks should be used.
        use_keypoints (bool): Indicates if keypoints should be used for pose estimation.
        use_obb (bool): Indicates if oriented bounding boxes should be used.
        data (dict): Dataset configuration dictionary.

    Methods:
        cache_labels: Cache dataset labels, check images and read shapes.
        get_labels: Return list of label dictionaries for YOLO training.
        get_label_files: Return companion label files for the dataset's images.
        verify_args: Return the per-image verification function and its arguments.
        result_to_label: Convert one verification result into a label dict.
        verify_labels: Check box/segment consistency of the loaded labels.
        build_transforms: Build and append transforms to the list.
        close_mosaic: Disable mosaic, copy_paste, mixup and cutmix augmentations and build transformations.
        update_labels_info: Update label format for different tasks.
        collate_fn: Collate data samples into batches.

    Examples:
        >>> dataset = YOLODataset(img_path="path/to/images", data={"names": {0: "person"}}, task="detect")
        >>> dataset.get_labels()
    NÚdetect©ÚdataÚtaskr1   údict | Noner2   Ústrc               ó<  •— |dk    | _         |dk    | _        |dk    | _        || _        | j                             dd¦  «        \  }}| j        r|dk    s|dvrt          d¦  «        ‚ t          ¦   «         j        |d	| j                             d	d
¦  «        i|¤Ž dS )ag  Initialize the YOLODataset.

        Args:
            data (dict, optional): Dataset configuration dictionary.
            task (str): Task type, one of 'detect', 'segment', 'pose', or 'obb'.
            *args (Any): Additional positional arguments for the parent class.
            **kwargs (Any): Additional keyword arguments for the parent class.
        ÚsegmentÚposeÚobbÚ	kpt_shape©r   r   r   >   é   é   z¦'kpt_shape' in data.yaml missing or incorrect. Should be a list with [number of keypoints, number of dims (2 for x,y or 3 for x,y,visible)], i.e. 'kpt_shape: [17, 3]'Úchannelsr<   N)Úuse_segmentsÚuse_keypointsÚuse_obbr1   ÚgetÚ
ValueErrorÚsuperÚ__init__)Úselfr1   r2   ÚargsÚkwargsÚnkptÚndimÚ	__class__s          €úh/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/ultralytics/data/dataset.pyrD   zYOLODataset.__init__X   s·   ø€ ð ! IÒ-ˆÔØ! Vš^ˆÔØ˜u’}ˆŒØˆŒ	Ø”Y—]’] ;°Ñ7Ô7‰
ˆˆdØÔð 	 4¨1¢9 9°¸FÐ0BÐ0BÝðiñô ð ð 	�‰ŒÔ˜$ÐP¨¬¯ª°zÀ1Ñ)EÔ)EÐPÈÐPÐPÐPÐPÐPó    ú./labels.cacheÚpathr   ÚreturnÚdictc           
     ó(  — dg i}ddddg f\  }}}}}| j         › d|j        |j        z  › d�}t          | j        ¦  «        }	t          t          ¦  «        5 }
|                      ¦   «         \  }}|
                     ||¬¦  «        }t          |||	¬¦  «        }|D ]ˆ}|  
                    |¦  «        \  }}}}}}||z  }||z  }||z  }||z  }|�|d                              |¦  «         |r|                     |¦  «         |› d|                      ||||¦  «        › �|_        Œ‰|                     ¦   «          ddd¦  «         n# 1 swxY w Y   |r't          j        d	                     |¦  «        ¦  «         |dk    rN| j        r!t'          | j         › d
|› dt(          › �¦  «        ‚t          j        | j         › d
|› dt(          › �¦  «         |                      ¦   «         |d<   |||||	f|d<   ||d<   |d         rt/          | j         ||t0          ¦  «         |S )aè  Cache dataset labels, check images and read shapes.

        This is the shared scanning skeleton for file-based datasets; subclasses customize it through the
        `get_label_files`, `get_cache_hash`, `verify_args`, `result_to_label` and `scan_summary` hooks instead of
        duplicating this method.

        Args:
            path (Path): Path where to save the cache file.

        Returns:
            (dict): Dictionary containing cached labels and related information.
        Úlabelsr   ú	Scanning ú...©ÚfuncÚiterable©ÚdescÚtotalNú ú
zNo labels found in ú. ÚhashÚresultsÚmsgs)ÚprefixÚparentÚstemÚlenÚim_filesr   r   Úverify_argsÚimapr   Úresult_to_labelÚappendÚscan_summaryrY   Úcloser   ÚinfoÚjoinÚaugmentrB   r    ÚwarningÚget_cache_hashr(   ÚDATASET_CACHE_VERSION)rE   rN   ÚxÚnmÚnfÚneÚncr`   rY   rZ   ÚpoolrV   rW   r_   ÚpbarÚresultÚlabelÚnm_fÚnf_fÚne_fÚnc_fÚmsgs                         rK   Úcache_labelszYOLODataset.cache_labelsm   s›  € ð �rˆNˆØ  ! Q¨¨2˜~ÑˆˆB��B˜Ø”+ÐDÐD¨¬°d´iÑ(?ÐDÐDÐDˆÝ�D”MÑ"Ô"ˆÝ�Ñ$Ô$ð 	¨Ø!×-Ò-Ñ/Ô/‰NˆD�(Ø—i’i T°H�iÑ=Ô=ˆGÝ˜ d°%Ð8Ñ8Ô8ˆDØð 
Jð 
J�Ø59×5IÒ5IÈ&Ñ5QÔ5QÑ2��t˜T 4¨¨sØ�d‘
�Ø�d‘
�Ø�d‘
�Ø�d‘
�ØÐ$Ø�h”K×&Ò& uÑ-Ô-Ð-Øð %Ø—K’K Ñ$Ô$Ð$Ø#ÐIÐI d×&7Ò&7¸¸BÀÀBÑ&GÔ&GÐIÐI�”	�	Ø�JŠJ‰LŒLˆLð	ð 	ð 	ñ 	ô 	ð 	ð 	ð 	ð 	ð 	ð 	øøøð 	ð 	ð 	ð 	ð" ð 	)ÝŒK˜Ÿ	š	 $™œÑ(Ô(Ð(Ø�Š7ˆ7ØŒ|ð XÝ  D¤KÐ!VÐ!VÀDÐ!VÐ!VÍHÐ!VÐ!VÑWÔWÐWÝŒN˜dœkÐPÐP¸dÐPÐPÅhÐPÐPÑQÔQÐQØ×'Ò'Ñ)Ô)ˆˆ&‰	Ø˜2˜r 2 uÐ,ˆˆ)‰Øˆˆ&‰	ØˆXŒ;ð 	QÝ# D¤K°°qÕ:OÑPÔPÐPØˆs   ÁC E Å EÅEú	list[str]c                óB   — t          | j        ¦  «        | _        | j        S )z¦Return the companion label files for the dataset's images, storing them on the instance.

        Returns:
            (list[str]): List of label file paths.
        )r$   re   Úlabel_files©rE   s    rK   Úget_label_fileszYOLODataset.get_label_filesœ   s   € õ +¨4¬=Ñ9Ô9ˆÔØÔÐrL   c                óÜ   — | j         t          | j        d         ¦  «        | j                             d¦  «        | j        f}t          | j        | j        z   t          |¦  «        gz   ¦  «        S )z¥Return the hash used to validate a label cache against the current dataset files and scan settings.

        Returns:
            (str): Dataset cache hash.
        Únamesr9   )	r?   rd   r1   rA   Ú
single_clsr"   rƒ   re   r4   )rE   Ú	scan_argss     rK   rp   zYOLODataset.get_cache_hash¥   s]   € ð Ô'­¨T¬Y°wÔ-?Ñ)@Ô)@À$Ä)Ç-Â-ÐP[ÑB\ÔB\Ð^bÔ^mÐnˆ	Ý˜Ô(¨4¬=Ñ8½CÀ	¹N¼NÐ;KÑKÑLÔLÐLrL   rt   Úintrs   ru   rv   c                ó   — |› d||z   › d|› d�S )a[  Return a one-line summary of scan counters for progress bars and cache logs.

        Args:
            nf (int): Number of found images.
            nm (int): Number of missing labels.
            ne (int): Number of empty labels.
            nc (int): Number of corrupt images.

        Returns:
            (str): Scan summary message.
        ú	 images, z backgrounds, ú corrupt© ©rE   rt   rs   ru   rv   s        rK   rj   zYOLODataset.scan_summary®   s'   € ð ÐBÐB˜r B™wÐBÐB°bÐBÐBÐBÐBrL   Útuplec                óv  — | j                              dd¦  «        \  }}t          t          | j        | j        t          | j        ¦  «        t          | j        ¦  «        t          t          | j         d         ¦  «        ¦  «        t          |¦  «        t          |¦  «        t          | j
        ¦  «        ¦  «        fS )zÍReturn the per-image verification function and its argument iterable used by `cache_labels`.

        Returns:
            (tuple): (verify function, zipped argument iterable) for ThreadPool.imap.
        r9   r:   r‡   )r1   rA   r+   Úzipre   rƒ   r   ra   r?   rd   rˆ   )rE   rH   rI   s      rK   rf   zYOLODataset.verify_args¼   s•   € ð ”Y—]’] ;°Ñ7Ô7‰
ˆˆdÝ!¥3ØŒMØÔÝ�4”;ÑÔÝ�4Ô%Ñ&Ô&Ý•3�t”y Ô)Ñ*Ô*Ñ+Ô+Ý�4‰LŒLÝ�4‰LŒLÝ�4”?Ñ#Ô#ñ	$
ô 	$
ð 	
ð 		
rL   ry   Úlistú+tuple[dict | None, int, int, int, int, str]c           
     óz   — |\
  }}}}}}}}	}
}|r#|||dd…dd…f         |dd…dd…f         ||dddœnd}||||	|
|fS )a,  Convert one verification result into a label dict and scan counter increments.

        Args:
            result (list): One result from the verification function returned by `verify_args`.

        Returns:
            (tuple): (label dict or None, missing, found, empty, corrupt, message).
        Nr   r   TÚxywh)Úim_fileÚshapeÚclsÚbboxesÚsegmentsÚ	keypointsÚ
normalizedÚbbox_formatrŽ   )rE   ry   r—   Úlbr˜   r›   Úkeypointr{   r|   r}   r~   r   rz   s                rK   rh   zYOLODataset.result_to_labelÎ   s�   € ð OUÑKˆ��U˜H h°°d¸DÀ$Èð ðØ"ØØ˜!˜!˜!˜Q˜q˜S˜&”zØ˜Q˜Q˜Q   ˜Uœ)Ø$Ø%Ø"Ø%ð	ð 	ð 	ð ð 	ð �d˜D $¨¨cÐ1Ð1rL   rR   ú
list[dict]Ú
cache_pathÚNonec                ó:  — d„ |D ¦   «         }d„ t          |Ž D ¦   «         \  }}}| j        r||k    rt          d|› d|› d�¦  «        ‚|r+||k    r%t          j        d|› d|› d�¦  «         |D ]}g |d<   Œ|d	k    r!t          j        d
|› dt
          › �¦  «         dS dS )a  Check that the dataset is all boxes or all segments, removing mixed segments if necessary.

        Args:
            labels (list[dict]): List of label dictionaries.
            cache_path (Path): Path of the dataset cache file, used in warning messages.
        c              3  ó’   K  — | ]B}t          |d          ¦  «        t          |d         ¦  «        t          |d         ¦  «        fV — ŒCdS )r™   rš   r›   N©rd   ©Ú.0rŸ   s     rK   ú	<genexpr>z,YOLODataset.verify_labels.<locals>.<genexpr>ð   sL   è è € Ð]Ð]ÐPR•C˜˜5œ	‘N”N¥C¨¨8¬Ñ$5Ô$5µs¸2¸j¼>Ñ7JÔ7JÐKÐ]Ð]Ð]Ð]Ð]Ð]rL   c              3  ó4   K  — | ]}t          |¦  «        V — Œd S ©N)Úsum©r¨   rr   s     rK   r©   z,YOLODataset.verify_labels.<locals>.<genexpr>ñ   s(   è è € Ð+JÐ+J°q­C°©F¬FÐ+JÐ+JÐ+JÐ+JÐ+JÐ+JrL   zVSegment dataset requires equal numbers of boxes and segments, but got len(segments) = z, len(boxes) = z8. Please supply a segment dataset, not a detect dataset.z@Box and segment counts should be equal, but got len(segments) = z¯. To resolve this only boxes will be used and all segments will be removed. To avoid this please supply either a detect or segment dataset, not a detect-segment mixed dataset.r›   r   zLabels are missing or empty in z#, training may not work correctly. N)r’   r>   rB   r   ro   r    )rE   rR   r¢   ÚlengthsÚlen_clsÚ	len_boxesÚlen_segmentsrŸ   s           rK   Úverify_labelszYOLODataset.verify_labelsè   s.  € ð ^Ð]ÐV\Ð]Ñ]Ô]ˆØ+JÐ+J½CÀ¸MÐ+JÑ+JÔ+JÑ(ˆ�˜LØÔð 	 ¨lÒ!:Ð!:ÝðtØðtð tØ09ðtð tð tñô ð ð ð 	$˜I¨Ò5Ð5ÝŒNðvÐS_ð vð vØ )ðvð vð vñô ð ð
 ð $ð $�Ø!#��:‘�Ø�aŠ<ˆ<ÝŒNÐv¸ZÐvÐvÕltÐvÐvÑwÔwÐwÐwÐwð ˆ<rL   Ú
cache_hashútuple[dict, bool]c                óæ   — 	 t          |¦  «        d}}|d         t          k    sJ ‚|d         |k    sJ ‚n:# t          t          t          t
          f$ r |                      |¦  «        d}}Y nw xY w||fS )a\  Load a dataset cache file if it matches the current version and hash, otherwise rescan and rebuild it.

        Args:
            cache_path (Path): Path of the cache file.
            cache_hash (str): Expected hash of the dataset files.

        Returns:
            (tuple): (cache dict, True if a valid existing cache file was loaded).
        TÚversionr^   F)r%   rq   ÚFileNotFoundErrorÚAssertionErrorÚAttributeErrorÚModuleNotFoundErrorr€   )rE   r¢   r³   ÚcacheÚexistss        rK   Ú_load_or_scan_cachezYOLODataset._load_or_scan_cache  s–   € ð	AÝ3°JÑ?Ô?À�6ˆEØ˜Ô#Õ'<Ò<Ð<Ð<Ð<Ø˜”= JÒ.Ð.Ð.Ð.Ð.øÝ!¥>µ>ÕCVÐWð 	Að 	Að 	AØ ×-Ò-¨jÑ9Ô9¸5�6ˆEˆEˆEð	Aøøøà�fˆ}Ðs   ‚25 µ4A,Á+A,c           	     ó.  ‡— |                       ¦   «         }t          |d         ¦  «        j                             d¦  «        }|                      ||                      ¦   «         ¦  «        \  Š}‰                     d¦  «        \  }}}}}|rwt          dv rnd|› d|                      ||||¦  «        › �}	t          d| j
        |	z   ||¬¦  «         ‰d	         r-t          j        d
                     ‰d	         ¦  «        ¦  «         ‰d         }
|
sTd                     t          t          ‰d	         ¦  «        ¦  «        ¦  «        pd}t!          d|› d|› d
t"          › �¦  «        ‚ˆfd„dD ¦   «          d„ |
D ¦   «         | _        |                      |
|¦  «         |
S )a5  Return list of label dictionaries for YOLO training.

        This method loads labels from disk or cache, verifies their integrity, and prepares them for training.

        Returns:
            (list[dict]): List of label dictionaries, each containing information about an image and its annotations.
        r   ú.cacher_   ¾   r   éÿÿÿÿrS   z... N©rY   rZ   Úinitialr`   r\   rR   z
  zno error detailszNo valid images found in z.
  c                ó:   •— g | ]}‰                      |¦  «        ‘ŒS rŽ   ©Úpop©r¨   Úkr»   s     €rK   ú
<listcomp>z*YOLODataset.get_labels.<locals>.<listcomp>-  s#   ø€ Ð;Ð;Ð;˜!ˆ�Š�1‰ŒÐ;Ð;Ð;rL   )r^   r¶   r`   c                ó   — g | ]
}|d          ‘ŒS ©r—   rŽ   r§   s     rK   rÉ   z*YOLODataset.get_labels.<locals>.<listcomp>.  s   € Ð8Ð8Ð8¨2˜˜IœÐ8Ð8Ð8rL   )r…   r   rb   Úwith_suffixr½   rp   rÆ   r   rj   r   ra   r   rl   rm   ÚsortedÚsetÚRuntimeErrorr    re   r²   )rE   rƒ   r¢   r¼   rt   rs   ru   rv   ÚnÚdrR   Úissuesr»   s               @rK   Ú
get_labelszYOLODataset.get_labels  s±  ø€ ð ×*Ò*Ñ,Ô,ˆÝ˜+ aœ.Ñ)Ô)Ô0×<Ò<¸XÑFÔFˆ
Ø×0Ò0°¸T×=PÒ=PÑ=RÔ=RÑSÔS‰ˆˆvð "ŸIšI iÑ0Ô0ÑˆˆB��B˜Øð 	6•j GÐ+Ð+ØO˜JÐOÐO¨D×,=Ò,=¸bÀ"ÀbÈ"Ñ,MÔ,MÐOÐOˆAÝ�˜DœK¨!™O°1¸aÐ@Ñ@Ô@Ð@Ø�VŒ}ð 6Ý”˜DŸIšI e¨F¤mÑ4Ô4Ñ5Ô5Ð5ð �x”ˆØð 	bØ—[’[¥­¨E°&¬MÑ(:Ô(:Ñ!;Ô!;Ñ<Ô<ÐRÐ@RˆFÝÐ`¸:Ð`Ð`ÈFÐ`Ð`ÕV^Ð`Ð`ÑaÔaÐaØ;Ð;Ð;Ð;Ð:Ð;Ñ;Ô;Ð;Ø8Ð8°Ð8Ñ8Ô8ˆŒØ×Ò˜6 :Ñ.Ô.Ð.ØˆrL   Úhypr   c                ón  — | j         rª| j         r| j        s|j        nd|_        | j         r| j        s|j        nd|_        | j         r| j        s|j        nd|_        t          | | j        |¦  «        }| j        t          u r1t          |j
        p|j        pt          |dd¦  «        ¦  «        | _        n+t          t          | j        | j        fd¬¦  «        g¦  «        }|                     |                      dd| j        | j        | j        d|j        |j        | j         r|j        nd¬¦	  «	        ¦  «         |S )	zÀBuild and append transforms to the list.

        Args:
            hyp (dict, optional): Hyperparameters for transforms.

        Returns:
            (Compose): Composed transforms.
        ç        ÚaugmentationsNF)Ú	new_shapeÚscaleupr–   T)	rž   Ú	normalizeÚreturn_maskÚreturn_keypointÚ
return_obbÚ	batch_idxÚ
mask_ratioÚmask_overlapÚbgr)rn   ÚrectÚmosaicÚmixupÚcutmixr   ÚimgszÚformat_classr   ÚboolÚ
copy_pasteÚgetattrr>   r   r   ri   r?   r@   rß   Úoverlap_maskrá   )rE   rÔ   Ú
transformss      rK   Úbuild_transformszYOLODataset.build_transforms2  sC  € ð Œ<ð 	aØ'+¤|ÐN¸D¼IÐN˜œ˜È3ˆCŒJØ%)¤\ÐL¸$¼)ÐL˜œ	˜	ÈˆCŒIØ'+¤|ÐN¸D¼IÐN˜œ˜È3ˆCŒJÝ& t¨T¬Z¸Ñ=Ô=ˆJØÔ ¥NÐ2Ð2Ý$(¨¬Ð)l¸3¼:Ð)lÍÐQTÐVeÐgkÑIlÔIlÑ$mÔ$m�Ô!øå ¥)°t´zÀ4Ä:Ð6NÐX]Ð"^Ñ"^Ô"^Ð!_Ñ`Ô`ˆJØ×ÒØ×ÒØ"ØØ Ô-Ø $Ô 2Øœ<ØØœ>Ø Ô-Ø#œ|Ð4�C”G�G°ð ñ 
ô 
ñ	
ô 	
ð 	
ð ÐrL   rì   Úmax_samplesc                ó°   — | j         rNt          t          |d¦  «        d|                      | j        ¦  «        ¬¦  «        }|                     d|¦  «         |S )ai  Insert text augmentation for text-based subclasses providing `category_freq`.

        Args:
            transforms (Compose): Transforms composed by build_transforms.
            max_samples (int): Maximum number of text samples per image.

        Returns:
            (Compose): Transforms with RandomLoadText inserted before Format when augmenting.
        éP   T)rî   ÚpaddingÚpadding_valuerÁ   )rn   r   ÚminÚ_get_neg_textsÚcategory_freqÚinsert)rE   rì   rî   Ú	transforms       rK   Úbuild_text_transformsz!YOLODataset.build_text_transformsS  sf   € ð Œ<ð 
	-õ
 'Ý ¨RÑ0Ô0ØØ"×1Ò1°$Ô2DÑEÔEðñ ô ˆIð
 ×Ò˜b )Ñ,Ô,Ð,ØÐrL   rõ   c                ó¢   ‡— t          t          |                      ¦   «         ¦  «        d¦  «        Šˆfd„|                      ¦   «         D ¦   «         S )zEGet negative text samples with frequency above the dataset threshold.éd   c                ó&   •— g | ]\  }}|‰k    ¯|‘ŒS rŽ   rŽ   )r¨   rÈ   ÚvÚ	thresholds      €rK   rÉ   z.YOLODataset._get_neg_texts.<locals>.<listcomp>n  s"   ø€ ÐFÐFÐF‘d�a˜°q¸I²~°~�°~°~°~rL   )ró   ÚmaxÚvaluesÚitems)rõ   rý   s    @rK   rô   zYOLODataset._get_neg_textsj  sN   ø€ õ �˜M×0Ò0Ñ2Ô2Ñ3Ô3°SÑ9Ô9ˆ	ØFÐFÐFÐF˜m×1Ò1Ñ3Ô3ÐFÑFÔFÐFrL   c                ór   — d|_         d|_        d|_        d|_        |                      |¦  «        | _        dS )úªDisable mosaic, copy_paste, mixup and cutmix augmentations by setting their values to 0.0.

        Args:
            hyp (dict): Hyperparameters for transforms.
        rÖ   N)rã   ré   rä   rå   rí   rì   )rE   rÔ   s     rK   Úclose_mosaiczYOLODataset.close_mosaicp  s9   € ð ˆŒ
ØˆŒØˆŒ	ØˆŒ
Ø×/Ò/°Ñ4Ô4ˆŒˆˆrL   rz   c                óN  — |                      d¦  «        }|                      dg ¦  «        }|                      dd¦  «        }|                      d¦  «        }|                      d¦  «        }| j        rdnd}t          |¦  «        d	k    ra| j        s| j        t
          urLt          d
„ |D ¦   «         ¦  «        }||k     r|dz   n|}t          j        t          ||¬¦  «        d	¬¦  «        }n#t          j
        d	|dft          j        ¬¦  «        }t          |||||¬¦  «        |d<   |S )aÓ  Update label format for different tasks.

        Args:
            label (dict): Label dictionary containing bboxes, segments, keypoints, etc.

        Returns:
            (dict): Updated label dictionary with instances.

        Notes:
            cls is not with bboxes now, classification and semantic segmentation need an independent cls label
            Can also support classification and semantic segmentation by adding or removing dict keys there.
        rš   r›   rœ   Nrž   r�   rú   éè  r   c              3  ó4   K  — | ]}t          |¦  «        V — Œd S r«   r¦   )r¨   Úss     rK   r©   z1YOLODataset.update_labels_info.<locals>.<genexpr>“  s(   è è € Ð3Ð3 Q�#˜a™&œ&Ð3Ð3Ð3Ð3Ð3Ð3rL   r   )rÐ   ©Úaxisr;   ©Údtype)rž   r�   Ú	instances)rÆ   r@   rd   r>   rç   r   rþ   ÚnpÚstackr   ÚzerosÚfloat32r   )	rE   rz   rš   r›   rœ   rž   r�   Úsegment_resamplesÚmax_lens	            rK   Úupdate_labels_infozYOLODataset.update_labels_info|  s4  € ð —’˜8Ñ$Ô$ˆØ—9’9˜Z¨Ñ,Ô,ˆØ—I’I˜k¨4Ñ0Ô0ˆ	Ø—i’i Ñ.Ô.ˆØ—Y’Y˜|Ñ,Ô,ˆ
ð $(¤<Ð9˜C˜C°TÐÝˆx‰=Œ=˜1ÒÐ $Ô"3Ð°tÔ7HÕP^Ð7^Ð7^åÐ3Ð3¨(Ð3Ñ3Ô3Ñ3Ô3ˆGØ1BÀWÒ1LÐ1L ¨1¡ ÐRcÐå”xÕ 1°(Ð>OÐ PÑ PÔ PÐWXÐYÑYÔYˆHˆHå”x Ð$5°qÐ 9ÅÄÐLÑLÔLˆHÝ& v¨x¸ÐP[ÐhrÐsÑsÔsˆˆkÑØˆrL   Úbatchc                óŒ  — i }d„ | D ¦   «         } | d                               ¦   «         }t          t          d„ | D ¦   «         Ž ¦  «        }t          |¦  «        D ]v\  }}||         }|dv rt	          j        |d¦  «        }n1|dk    r+t          j        j        j         	                    |d¬¦  «        }|dv rt	          j
        |d¦  «        }|||<   Œwd	|v rqt          |d	         ¦  «        |d	<   t          t          |d	         ¦  «        ¦  «        D ]}|d	         |xx         |z  cc<   Œt	          j
        |d	         d¦  «        |d	<   |S )
úÒCollate data samples into batches.

        Args:
            batch (list[dict]): List of dictionaries containing sample data.

        Returns:
            (dict): Collated batch with stacked tensors.
        c                ój   — g | ]0}t          t          |                     ¦   «         ¦  «        ¦  «        ‘Œ1S rŽ   )rP   rÍ   r   ©r¨   Úbs     rK   rÉ   z*YOLODataset.collate_fn.<locals>.<listcomp>§  s0   € Ð8Ð8Ð8¨Q••f˜QŸWšW™YœYÑ'Ô'Ñ(Ô(Ð8Ð8Ð8rL   r   c                óP   — g | ]#}t          |                     ¦   «         ¦  «        ‘Œ$S rŽ   )r“   rÿ   r  s     rK   rÉ   z*YOLODataset.collate_fn.<locals>.<listcomp>©  s(   € Ð<Ð<Ð<°�D §¢¡¤Ñ,Ô,Ð<Ð<Ð<rL   >   ÚimgÚdepthÚ	sem_masksÚ
text_featsÚsemantic_maskÚvisualsT)Úbatch_first>   r™   r8   Úmasksrš   r›   rœ   rÞ   )Úkeysr“   r’   Ú	enumerateÚtorchr  ÚnnÚutilsÚrnnÚpad_sequenceÚcatÚrangerd   )r  Ú	new_batchr#  rÿ   ÚirÈ   Úvalues          rK   Ú
collate_fnzYOLODataset.collate_fnœ  sj  € ð ˆ	Ø8Ð8°%Ð8Ñ8Ô8ˆØ�QŒx�}Š}‰ŒˆÝ•cÐ<Ð<°eÐ<Ñ<Ô<Ð=Ñ>Ô>ˆÝ˜d‘O”Oð 	!ð 	!‰DˆAˆqØ˜1”IˆEØÐPÐPÐPÝœ E¨1Ñ-Ô-��Ø�i’�ÝœœÔ*×7Ò7¸È4Ð7ÑPÔP�ØÐNÐNÐNÝœ	 %¨Ñ+Ô+�Ø ˆI�a‰LˆLØ˜)Ð#Ð#Ý%)¨)°KÔ*@Ñ%AÔ%AˆI�kÑ"Ý�3˜y¨Ô5Ñ6Ô6Ñ7Ô7ð /ð /�Ø˜+Ô& qÐ)Ð)Ô)¨QÑ.Ð)Ð)Ñ)Ð)Ý%*¤Y¨y¸Ô/EÀqÑ%IÔ%IˆI�kÑ"ØÐrL   ©r1   r3   r2   r4   )rN   r   rO   rP   ©rO   r�   ©rO   r4   ©
rt   rŠ   rs   rŠ   ru   rŠ   rv   rŠ   rO   r4   ©rO   r�   )ry   r“   rO   r”   ©rR   r¡   r¢   r   rO   r£   )r¢   r   r³   r4   rO   r´   ©rO   r¡   r«   ©rÔ   r3   rO   r   )rì   r   rî   rŠ   rO   r   )rõ   rP   rO   r�   ©rÔ   rP   rO   r£   ©rz   rP   rO   rP   ©r  r¡   rO   rP   )Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   rç   rD   r   r€   r…   rp   rj   rf   rh   r²   r½   rÓ   rí   rø   Ústaticmethodrô   r  r  r/  Ú__classcell__©rJ   s   @rK   r.   r.   8   sÖ  ø€ € € € € ðð ð: €Là26ÀHð Qð Qð Qð Qð Qð Qð Qð Qð* )-¨Ð-=Ñ(>Ô(>ð -ð -ð -ð -ð -ð^ ð  ð  ð  ðMð Mð Mð MðCð Cð Cð Cð
ð 
ð 
ð 
ð$2ð 2ð 2ð 2ð4xð xð xð xð4ð ð ð ð$ð ð ð ð<ð ð ð ð ðBð ð ð ð. ðGð Gð Gñ „\ðGð

5ð 
5ð 
5ð 
5ðð ð ð ð@ ðð ð ñ „\ðð ð ð ð rL   r.   c                  óp   ‡ — e Zd ZdZeZdd„Zdd„Zd d„Zd!d„Z	d"d„Z
d#d„Zd$d„Zd„ Zˆ fd„Zd%ˆ fd„	Zˆ xZS )&ÚDepthDatasetu†  Dataset for monocular depth estimation with paired RGB + depth map loading.

    Extends YOLODataset to load depth ground truth maps alongside RGB images. Depth maps are stored as PNG or NPY files
    in a parallel directory structure (images/train/*.jpg â†’ depth/train/*.{png,npy}).

    Examples:
        >>> dataset = DepthDataset(img_path="/data/nyu/images/train", data={"nc": 1})
    r—   r4   rO   c                óZ  — t          t          |¦  «        j        ¦  «        }t          t	          |¦  «        dz
  dd¦  «        D ]}||         dk    rd||<    nŒt          |Ž                      d¦  «        }t          |                     ¦   «         r|n|                     d¦  «        ¦  «        S )z;Map an image path to its companion PNG or NPY depth target.r   rÁ   Úimagesr  ú.pngú.npy)r“   r   Úpartsr+  rd   rÌ   r4   Úis_file)rE   r—   rH  r-  rN   s        rK   Ú_depth_path_forzDepthDataset._depth_path_forÇ  s¥   € å•T˜'‘]”]Ô(Ñ)Ô)ˆÝ•s˜5‘z”z A‘~ r¨2Ñ.Ô.ð 	ð 	ˆAØ�QŒx˜8Ò#Ð#Ø"��a‘Ø�ð $õ �Uˆ|×'Ò'¨Ñ/Ô/ˆÝ˜4Ÿ<š<™>œ>ÐG�4�4¨t×/?Ò/?ÀÑ/GÔ/GÑHÔHÐHrL   r�   c                ó˜   ‡ — ˆ fd„‰ j         D ¦   «         ‰ _        t          ‰ j                             ¦   «         ¦  «        ‰ _        ‰ j        S )z†Return the depth paths paired with the dataset's images.

        Returns:
            (list[str]): List of depth file paths.
        c                ó<   •— i | ]}|‰                      |¦  «        “ŒS rŽ   )rJ  )r¨   ÚfrE   s     €rK   ú
<dictcomp>z0DepthDataset.get_label_files.<locals>.<dictcomp>×  s)   ø€ Ð$WÐ$WÐ$WÀA Q¨×(<Ò(<¸QÑ(?Ô(?Ð$WÐ$WÐ$WrL   )re   Údepth_files_by_imager“   rÿ   Údepth_filesr„   s   `rK   r…   zDepthDataset.get_label_filesÑ  sL   ø€ ð %XÐ$WÐ$WÐ$WÈÌÐ$WÑ$WÔ$WˆÔ!Ý Ô 9× @Ò @Ñ BÔ BÑCÔCˆÔØÔÐrL   c           	     óŽ   — t          | j        | j        z   t          | j                             dd¦  «        ¦  «        gz   ¦  «        S )zvReturn a hash over the paired depth and image files.

        Returns:
            (str): Dataset cache hash.
        Údepth_scaler  )r"   rP  re   r4   r1   rA   r„   s    rK   rp   zDepthDataset.get_cache_hashÛ  s=   € õ ˜Ô(¨4¬=Ñ8½CÀÄ	ÇÂÈmÐ]aÑ@bÔ@bÑ<cÔ<cÐ;dÑdÑeÔeÐerL   rt   rŠ   rs   ru   rv   c                ó   — |› d|› d|› d�S )z7Return a one-line summary of image-depth scan counters.rŒ   z missing depth, r�   rŽ   r�   s        rK   rj   zDepthDataset.scan_summaryã  ó!   € àÐ?Ð?˜rÐ?Ð?°2Ð?Ð?Ð?Ð?rL   r�   c                ó¸   — t           t          | j        | j        t	          | j        ¦  «        t	          | j                             dd¦  «        ¦  «        ¦  «        fS )zAReturn the depth verification function and its argument iterable.rR  r  )r*   r’   re   rP  r   ra   r1   rA   r„   s    rK   rf   zDepthDataset.verify_argsç  sM   € å!¥3ØŒM˜4Ô+­V°D´KÑ-@Ô-@Å&ÈÌÏÊÐWdÐfjÑIkÔIkÑBlÔBlñ$
ô $
ð 
ð 	
rL   ry   r”   c                ó¸   — |\  }}}}}}|rF||t          j        g t           j        ¬¦  «        t          j        dt           j        ¬¦  «        g dddœnd}|||d||fS )zTConvert one verify_image_depth result into a label dict and scan counter increments.r
  ©r   é   Tr–   )r—   r˜   r™   rš   r›   r�   rž   Nr   ©r  Úarrayr  r  )	rE   ry   r—   r˜   r|   r{   r~   r   rz   s	            rK   rh   zDepthDataset.result_to_labelí  sƒ   € à06Ñ-ˆ�˜˜d D¨#ð ð
Ø"ØÝ”x ­"¬*Ð5Ñ5Ô5Ýœ( 6µ´Ð<Ñ<Ô<ØØ"Ø%ðð ð ð ð 	ð �d˜D ! T¨3Ð.Ð.rL   rR   r¡   r¢   r   r£   c                ó   — dS )zPSkip box and segment checks; depth datasets carry no box or segment annotations.NrŽ   ©rE   rR   r¢   s      rK   r²   zDepthDataset.verify_labelsÿ  ó   € € € rL   c                ó€   — t          | j        | j        |                  | j                             dd¦  «        ¦  «        S )z4Return the native-resolution depth map for an image.rR  r  )r&   rO  re   r1   rA   )rE   Úindexs     rK   Ú_load_depthzDepthDataset._load_depth  s3   € å˜$Ô3°D´MÀ%Ô4HÔIÈ4Ì9Ï=Ê=ÐYfÐhlÑKmÔKmÑnÔnÐnrL   c                ó  •— t          ¦   «                              |¦  «        }|d         \  }}|                      |¦  «        }|j        dd…         ||fk    r#t	          j        |||ft          j        ¬¦  «        }||d<   |S )z5Load image, label, and depth map for the given index.Úresized_shapeNr;   ©Úinterpolationr  )rC   Úget_image_and_labelr`  r˜   Úcv2ÚresizeÚINTER_NEAREST)rE   r_  rz   ÚhÚwr  rJ   s         €rK   re  z DepthDataset.get_image_and_label  s   ø€ å‘”×+Ò+¨EÑ2Ô2ˆØ�_Ô%‰ˆˆ1Ø× Ò  Ñ'Ô'ˆØŒ;�r˜�rŒ?˜q !˜fÒ$Ð$Ý”J˜u q¨! f½CÔ<MÐNÑNÔNˆEØˆˆg‰ØˆrL   Nc                óÐ   •— dx|_         x|_        x|_        |_        t	          ¦   «                              |¦  «        }| j        s t          | j        | j        fd¬¦  «        |d<   |S )z¥Build transforms for depth estimation.

        Args:
            hyp (dict): Hyperparameters.

        Returns:
            (Compose): Composed transforms.
        rÖ   T)rØ   Ú
scale_filléþÿÿÿ)	rã   rä   rå   ré   rC   rí   rn   r   ræ   )rE   rÔ   rì   rJ   s      €rK   rí   zDepthDataset.build_transforms  sk   ø€ ð @CÐBˆŒ
ÐB�S”YÐB ¤¨c¬nÝ‘W”W×-Ò-¨cÑ2Ô2ˆ
ØŒ|ð 	\å&°$´*¸d¼jÐ1IÐVZÐ[Ñ[Ô[ˆJ�r‰NØÐrL   )r—   r4   rO   r4   r1  r2  r3  r4  ©ry   r�   rO   r”   r5  r«   )r;  r<  r=  r>  r   rç   rJ  r…   rp   rj   rf   rh   r²   r`  re  rí   r@  rA  s   @rK   rC  rC  »  s  ø€ € € € € ðð ð €LðIð Ið Ið Ið ð  ð  ð  ðfð fð fð fð@ð @ð @ð @ð
ð 
ð 
ð 
ð/ð /ð /ð /ð$_ð _ð _ð _ðoð oð oðð ð ð ð ðð ð ð ð ð ð ð ð ð rL   rC  c                  ór   ‡ — e Zd ZdZdddœdˆ fd	„Zdˆ fd„Zddˆ fd„Zed„ ¦   «         Zed„ ¦   «         Z	ˆ xZ
S )ÚYOLOMultiModalDatasetaœ  Dataset class for loading object detection and/or segmentation labels in YOLO format with multi-modal support.

    This class extends YOLODataset to add text information for multi-modal model training, enabling models to process
    both image and text data.

    Methods:
        update_labels_info: Add text information for multi-modal model training.
        build_transforms: Enhance data transformations with text augmentation.

    Examples:
        >>> dataset = YOLOMultiModalDataset(img_path="path/to/images", data={"names": {0: "person"}}, task="detect")
        >>> batch = next(iter(dataset))
        >>> print(batch.keys())  # Should include 'texts'
    Nr/   r0   r1   r3   r2   r4   c               ó@   •—  t          ¦   «         j        |||dœ|¤Ž dS )ao  Initialize a YOLOMultiModalDataset.

        Args:
            data (dict, optional): Dataset configuration dictionary.
            task (str): Task type, one of 'detect', 'segment', 'pose', or 'obb'.
            *args (Any): Additional positional arguments for the parent class.
            **kwargs (Any): Additional keyword arguments for the parent class.
        r0   N)rC   rD   )rE   r1   r2   rF   rG   rJ   s        €rK   rD   zYOLOMultiModalDataset.__init__2  s.   ø€ ð 	�‰ŒÔ˜$ T°Ð?Ð?¸Ð?Ð?Ð?Ð?Ð?rL   rz   rP   rO   c                ó¢   •— t          ¦   «                              |¦  «        }d„ | j        d                              ¦   «         D ¦   «         |d<   |S )zýAdd text information for multi-modal model training.

        Args:
            label (dict): Label dictionary containing bboxes, segments, keypoints, etc.

        Returns:
            (dict): Updated label dictionary with instances and texts.
        c                ó>   — g | ]\  }}|                      d ¦  «        ‘ŒS ©ú/©Úsplit)r¨   Ú_rü   s      rK   rÉ   z<YOLOMultiModalDataset.update_labels_info.<locals>.<listcomp>I  s&   € ÐOÐOÐO©D¨A¨q˜1Ÿ7š7 3™<œ<ÐOÐOÐOrL   r‡   Útexts)rC   r  r1   r   )rE   rz   rR   rJ   s      €rK   r  z(YOLOMultiModalDataset.update_labels_info=  sN   ø€ õ ‘”×+Ò+¨EÑ2Ô2ˆð PÐO°D´I¸gÔ4F×4LÒ4LÑ4NÔ4NÐOÑOÔOˆˆw‰àˆrL   rÔ   r   c                ó„   •— |                       t          ¦   «                              |¦  «        | j        d         ¦  «        S )a  Enhance data transformations with text augmentation for multi-modal training.

        Args:
            hyp (dict, optional): Hyperparameters for transforms.

        Returns:
            (Compose): Composed transforms including text augmentation if applicable.
        rv   )rø   rC   rí   r1   ©rE   rÔ   rJ   s     €rK   rí   z&YOLOMultiModalDataset.build_transformsM  s4   ø€ ð ×)Ò)­%©'¬'×*BÒ*BÀ3Ñ*GÔ*GÈÌÐSWÌÑYÔYÐYrL   c                óX   — | j         d                              ¦   «         }d„ |D ¦   «         S )zmReturn category names for the dataset.

        Returns:
            (set[str]): Set of class names.
        r‡   c                óf   — h | ].}|                      d ¦  «        D ]}|                     ¦   «         ’ŒŒ/S rt  )rw  Ústrip)r¨   ÚnamerÐ   s      rK   ú	<setcomp>z7YOLOMultiModalDataset.category_names.<locals>.<setcomp>`  s7   € ÐEÐEÐE˜d°T·Z²ZÀ±_´_ÐEÐE°�—’‘	”	ÐEÐEÐEÐErL   )r1   rÿ   )rE   r‡   s     rK   Úcategory_namesz$YOLOMultiModalDataset.category_namesX  s0   € ð ”	˜'Ô"×)Ò)Ñ+Ô+ˆØEÐE eÐEÑEÔEÐErL   c                ó–  — d„ | j         d                              ¦   «         D ¦   «         }t          t          ¦  «        }| j        D ]^}|d                              d¦  «        D ]@}|t          |¦  «                 }|D ]&}|                     ¦   «         }||xx         dz  cc<   Œ'ŒAŒ_|p$t                               d„ |D ¦   «         d¦  «        S )ú1Return frequency of each category in the dataset.c                ó8   — g | ]}|                      d ¦  «        ‘ŒS rt  rv  )r¨   rü   s     rK   rÉ   z7YOLOMultiModalDataset.category_freq.<locals>.<listcomp>e  s"   € ÐCÐCÐC !�—’˜‘”ÐCÐCÐCrL   r‡   r™   rÁ   r   c              3  óH   K  — | ]}|D ]}|                      ¦   «         V — ŒŒd S r«   ©r~  )r¨   ÚtextÚts      rK   r©   z6YOLOMultiModalDataset.category_freq.<locals>.<genexpr>n  s8   è è € Ð.YÐ.Y¸TÐTXÐ.YÐ.YÈq¨q¯wªw©y¬yÐ.YÐ.YÐ.YÐ.YÐ.YÐ.YÐ.YrL   r   )	r1   rÿ   r   rŠ   rR   Úsqueezer~  rP   Úfromkeys)rE   ry  rõ   rz   Úcr‡  rˆ  s          rK   rõ   z#YOLOMultiModalDataset.category_freqb  sé   € ð DÐC t¤y°Ô'9×'@Ò'@Ñ'BÔ'BÐCÑCÔCˆÝ#¥CÑ(Ô(ˆØ”[ð 	*ð 	*ˆEØ˜5”\×)Ò)¨"Ñ-Ô-ð *ð *�Ø�S ™VœV”}�Øð *ð *�AØŸš™	œ	�AØ! !Ð$Ð$Ô$¨Ñ)Ð$Ð$Ñ$Ð$ð*ð*ð Ð]¥§¢Ð.YÐ.YÀUÐ.YÑ.YÔ.YÐ[\Ñ ]Ô ]Ð]rL   r0  r9  r«   r7  )r;  r<  r=  r>  rD   r  rí   Úpropertyr�  rõ   r@  rA  s   @rK   rp  rp  "  sè   ø€ € € € € ðð ð 37ÀHð 	@ð 	@ð 	@ð 	@ð 	@ð 	@ð 	@ð 	@ðð ð ð ð ð ð 	Zð 	Zð 	Zð 	Zð 	Zð 	Zð 	Zð ðFð Fñ „XðFð ð^ð ^ñ „Xð^ð ^ð ^ð ^ð ^rL   rp  c                  óª   ‡ — e Zd ZdZddddœd#ˆ fd„Zd$ˆ fd„Zd%d„Zd&d„Z ed¦  «        fd'd„Z	d(d„Z
d)d*ˆ fd „Zed!„ ¦   «         Zed"„ ¦   «         Zˆ xZS )+ÚGroundingDatasetaÒ  Dataset class for object detection tasks using annotations from a JSON file in grounding format.

    This dataset is designed for grounding tasks where annotations are provided in a JSON file rather than the standard
    YOLO format text files.

    Attributes:
        json_file (str): Path to the JSON file containing annotations.

    Methods:
        get_labels: Load annotations from a JSON file and prepare them for training.
        build_transforms: Configure augmentations for training with optional text loading.

    Examples:
        >>> dataset = GroundingDataset(img_path="path/to/images", json_file="annotations.json", task="detect")
        >>> len(dataset)  # Number of valid images with annotations
    r/   Ú rð   )r2   Ú	json_filerî   r2   r4   r�  rî   rŠ   c               ó|   •— |dv s
J d¦   «         ‚|| _         || _         t          ¦   «         j        ||ddidœ|¤Ž dS )aØ  Initialize a GroundingDataset for object detection.

        Args:
            json_file (str): Path to the JSON file containing annotations.
            task (str): Must be 'detect' or 'segment' for GroundingDataset.
            max_samples (int): Maximum number of samples to load for text augmentation.
            *args (Any): Additional positional arguments for the parent class.
            **kwargs (Any): Additional keyword arguments for the parent class.
        >   r/   r6   zEGroundingDataset currently only supports `detect` and `segment` tasksr=   r<   )r2   r1   N)r�  rî   rC   rD   )rE   r2   r�  rî   rF   rG   rJ   s         €rK   rD   zGroundingDataset.__init__ƒ  s\   ø€ ð Ð,Ð,Ð,Ð,Ð.uÑ,Ô,Ð,Ø"ˆŒØ&ˆÔØ�‰ŒÔ˜$ T°¸Q°ÐJÐJÀ6ÐJÐJÐJÐJÐJrL   Úimg_pathrO   r�   c                ól   •— d| _         t          ¦   «                              |¦  «        | _        | j        S )zaReturn every image under `img_path`; the annotations, not `fraction`, decide which ones are used.ç      ð?)ÚfractionrC   Úget_img_filesÚ
scan_files)rE   r’  rJ   s     €rK   r–  zGroundingDataset.get_img_files’  s+   ø€ àˆŒÝ™'œ'×/Ò/°Ñ9Ô9ˆŒØŒÐrL   c                ó:   — t          | j        g| j        ¢¦  «        S )zEReturn a hash over the annotation file and images scanned against it.)r"   r�  r—  r„   s    rK   rp   zGroundingDataset.get_cache_hash˜  s   € å˜œÐ:¨$¬/Ð:Ñ;Ô;Ð;rL   rR   úlist[dict[str, Any]]r£   c           	     ó
  — dddddœ}t          d„ |D ¦   «         ¦  «        }|                     ¦   «         D ]0\  }}|| j        v r"||k    sJ d| j        › d|› d	|› d
�¦   «         ‚ dS Œ1t          j        d| j        › d�¦  «         dS )z4Verify instance counts for known grounding datasets.iLâ7 iÓ+8 i½	 iÀÆ	 )Úfinal_mixed_train_no_coco_segmÚfinal_mixed_train_no_cocoÚ"final_flickr_separateGT_train_segmÚfinal_flickr_separateGT_trainc              3  ó<   K  — | ]}|d          j         d         V — ŒdS )rš   r   N)r˜   ©r¨   rz   s     rK   r©   z;GroundingDataset._verify_instance_counts.<locals>.<genexpr>¥  s.   è è € ÐJÐJ¸%˜U 8œ_Ô2°1Ô5ÐJÐJÐJÐJÐJÐJrL   ú'z' has z instances, expected ú.Nz?Skipping instance count verification for unrecognized dataset ')r¬   r   r�  r   ro   )rE   rR   Úexpected_countsÚinstance_countÚ	data_nameÚcounts         rK   Ú_verify_instance_countsz(GroundingDataset._verify_instance_countsœ  sÓ   € ð /6Ø)0Ø28Ø-3ð	
ð 
ˆõ ÐJÐJÀ6ÐJÑJÔJÑJÔJˆØ /× 5Ò 5Ñ 7Ô 7ð 	ð 	ÑˆI�uØ˜DœNÐ*Ð*Ø%¨Ò.Ð.Ð.Ð0w°D´NÐ0wÐ0wÈ.Ð0wÐ0wÐotÐ0wÐ0wÐ0wÑ.Ô.Ð.Ø��ð +õ 	ŒÐjÐY]ÔYgÐjÐjÐjÑkÔkÐkÐkÐkrL   rM   rN   r   údict[str, Any]c                óJ  ‡$— dg i}t          j        d¦  «         t          | j        ¦  «        5 }t	          j        |¦  «        }ddd¦  «         n# 1 swxY w Y   d„ |d         D ¦   «         }t          t          ¦  «        }|d         D ]#}||d                                       |¦  «         Œ$d}t          | 
                    ¦   «         d	| j        › �¬
¦  «        D �]w\  }	}
||	d›         }|d         |d         |d         }}}t          | j        ¦  «        |z  }|                     ¦   «         sŒTg }g }d}i }g }|
D �]_}|d         rŒt          j        |d         t          j        ¬¦  «        }|dd…xx         |dd…         dz  z  cc<   |ddgxx         t#          |¦  «        z  cc<   |ddgxx         t#          |¦  «        z  cc<   |d         dk    s|d         dk    rŒ¦|d         Š$d                     ˆ$fd„|d         D ¦   «         ¦  «                             ¦   «                              ¦   «         }|sŒü||vr(t+          |¦  «        ||<   |                     |g¦  «         ||         }|g|                     ¦   «         ¢}||v�r|                     |¦  «         |                     d¦  «        }||duz  }t1          |t          ¦  «        r|ng }d„ |D ¦   «         }|t3          |¦  «        o5t1          |t          ¦  «         pt+          |¦  «        t+          |¦  «        k     z  }|sR|dd…         \  }}}}||dz  z
  ||dz  z
  ||dz  z   ||dz  z   f\  }}} }!|                     |||| || |!||!g	¦  «         �Œ>t+          |¦  «        dk    rnt5          |¦  «        }"t          j        |"d¬¦  «        t          j        ||gt          j        ¬¦  «        z                       d¦  «                             ¦   «         }"nˆd„ |D ¦   «         }"t          j        |"t          j        ¬¦  «                             dd¦  «        t          j        ||gt          j        ¬¦  «        z                       d¦  «                             ¦   «         }"|                     |g|"¢¦  «         �Œat+          |¦  «        r t          j        |t          j        ¬¦  «        nt          j        dt          j        ¬¦  «        }#|r%d „ |D ¦   «         }t=          |¦  «        |#dd…dd…f<   ng }|d                              |||f|#dd…dd…f         |#dd…dd…f         |d!d"|d#œ¦  «         �Œy|rt          j        | j        › d$�¦  «         |                       ¦   «         |d%<   tC          | j"        ||tF          ¦  «         |S )&a  Load annotations from a JSON file, filter, and normalize bounding boxes for each image.

        Args:
            path (Path): Path where to save the cache file.

        Returns:
            (dict[str, Any]): Dictionary containing cached labels and related information.
        rR   zLoading annotation file...Nc                ó$   — i | ]}|d          d›|“ŒS )ÚidrÑ   rŽ   r­   s     rK   rN  z1GroundingDataset.cache_labels.<locals>.<dictcomp>¹  s"   € ÐCÐCÐC¨�Q�t”W�.�. !ÐCÐCÐCrL   rE  r   Úimage_idFzReading annotations )rY   rÑ   ÚheightÚwidthÚ	file_nameÚiscrowdÚbboxr
  r;   r   r   r<   Úcaptionr[   c                ó<   •— g | ]}‰|d          |d         …         ‘ŒS )r   r   rŽ   )r¨   rˆ  r²  s     €rK   rÉ   z1GroundingDataset.cache_labels.<locals>.<listcomp>Ô  s*   ø€ Ð$ZÐ$ZÐ$Z¸a W¨Q¨q¬T°A°a´D¨[Ô%9Ð$ZÐ$ZÐ$ZrL   Útokens_positiveÚsegmentationc                ó¸   — g | ]W}t          |t          ¦  «        r@t          |¦  «        d k    ¯*t          |¦  «        dz  °<t          d„ |D ¦   «         ¦  «        ¯U|‘ŒXS )é   r;   c              3  óN   K  — | ] }t          |t          t          f¦  «        V — Œ!d S r«   )Ú
isinstancerŠ   Úfloat©r¨   r‹  s     rK   r©   z;GroundingDataset.cache_labels.<locals>.<listcomp>.<genexpr>è  s0   è è € ÐGÐGÀ¥
¨1­sµE¨lÑ ;Ô ;ÐGÐGÐGÐGÐGÐGrL   )r¹  r“   rd   Úall)r¨   Úps     rK   rÉ   z1GroundingDataset.cache_labels.<locals>.<listcomp>â  sv   € ð  ð  ð  àÝ% a­Ñ.Ô.ð õ   ™FœF ašK˜KÝ # A¡¤¨¡
ð (åÐGÐGÀQÐGÑGÔGÑGÔGð (ð ð (˜K˜KrL   r  rÁ   c                ó   — g | ]	}|D ]}|‘ŒŒ
S rŽ   rŽ   )r¨   r-  Újs      rK   rÉ   z1GroundingDataset.cache_labels.<locals>.<listcomp>ô  s%   € Ð<Ð<Ð< 1¸!Ð<Ð<°Q˜QÐ<Ð<Ð<Ð<rL   )r   é   c                ó†   — g | ]>}t          j        |d d…         t           j        ¬¦  «                             dd¦  «        ‘Œ?S )r   Nr
  rÁ   r;   )r  rZ  r  Úreshaper­   s     rK   rÉ   z1GroundingDataset.cache_labels.<locals>.<listcomp>þ  sB   € Ð_Ð_Ð_ÐQR�BœH Q q r r¤Uµ"´*Ð=Ñ=Ô=×EÒEÀbÈ!ÑLÔLÐ_Ð_Ð_rL   Tr–   )r—   r˜   r™   rš   r›   r�   rž   ry  z�: ignored segmentations that are not polygon point lists, such as RLE masks. Annotations left without a polygon use a segment shaped like their bounding box.r^   )$r   rl   Úopenr�  ÚjsonÚloadr   r“   ri   r   r   r   r’  r¼   r  rZ  r  rº  rm   Úlowerr~  rd   ÚtolistrA   r¹  rè   r   ÚconcatenaterÂ  r  r   ro   rp   r(   ra   rq   )%rE   rN   rr   rM  r   rE  Úimg_to_annsÚannÚdroppedÚimg_idÚannsr  ri  rj  r—   rš   r›   Ú	segmentedÚcat2idry  ÚboxÚcat_namer™   Úraw_segÚsegÚpolygonsÚcxÚcyÚbwÚbhÚx1Úy1Úx2Úy2r  rŸ   r²  s%                                       @rK   r€   zGroundingDataset.cache_labels¬  sd  ø€ ð �rˆNˆÝŒÐ0Ñ1Ô1Ð1Ý�$”.Ñ!Ô!ð 	' QÝœ) A™,œ,ˆKð	'ð 	'ð 	'ñ 	'ô 	'ð 	'ð 	'ð 	'ð 	'ð 	'ð 	'øøøð 	'ð 	'ð 	'ð 	'àCÐC¨[¸Ô-BÐCÑCÔCˆÝ!¥$Ñ'Ô'ˆØ˜}Ô-ð 	5ð 	5ˆCØ˜˜JœÔ(×/Ò/°Ñ4Ô4Ð4Ð4ØˆÝ  ×!2Ò!2Ñ!4Ô!4Ð;bÐRVÔR`Ð;bÐ;bÐcÑcÔcð P	ñ P	‰LˆF�DØ˜F˜˜Ô'ˆCØ˜(”m S¨¤\°3°{Ô3C�!ˆqˆAÝ˜4œ=Ñ)Ô)¨AÑ-ˆGØ—>’>Ñ#Ô#ð ØØˆFØˆHØˆIØˆFØˆEØð 1/ñ 1/�Ø�y”>ð ØÝ”h˜s 6œ{µ"´*Ð=Ñ=Ô=�Ø�B�Q�B��”˜3˜q˜r˜rœ7 Q™;Ñ&��‘Ø�Q˜�F��”�u Q™xœxÑ'��‘Ø�Q˜�F��”�u Q™xœxÑ'��‘Ø�q”6˜Q’;�; # a¤&¨A¢+ +Øà˜iœ.�ØŸ8š8Ð$ZÐ$ZÐ$ZÐ$ZÀ3ÐGXÔCYÐ$ZÑ$ZÔ$ZÑ[Ô[×aÒaÑcÔc×iÒiÑkÔk�Øð Øà 6Ð)Ð)Ý'*¨6¡{¤{�F˜8Ñ$Ø—L’L ( Ñ,Ô,Ð,Ø˜XÔ&�ØÐ*˜SŸZšZ™\œ\Ð*�Ø˜fÐ$Ñ$Ø—M’M #Ñ&Ô&Ð&Ø!Ÿgšg nÑ5Ô5�GØ °Ð!4Ñ4�IÝ%/°½Ñ%>Ô%>ÐF˜'˜'ÀB�Cð ð  à!$ð ñ  ô  �Hð �t G™}œ}ÐlµjÀÍ$Ñ6OÔ6OÐ2OÐ2kÕSVÐW_ÑS`ÔS`ÕcfÐgjÑckÔckÒSkÑl�GØ#ð Ø),¨Q¨R¨R¬™˜˜B  BØ)+¨b°1©f©°b¸2À¹6±kÀ2ÈÈQÉÁ;ÐPRÐUWÐZ[ÑU[ÑP[Ð)[™˜˜B  BØ Ÿš¨¨b°"°b¸"¸bÀ"ÀbÈ"Ð(MÑNÔNÐNÙ Ý˜X™œ¨Ò*Ð*Ý/°Ñ9Ô9˜Ýœ^¨A°AÐ6Ñ6Ô6½¼À1ÀaÀ&ÕPRÔPZÐ9[Ñ9[Ô9[Ñ[×dÒdÐegÑhÔh×oÒoÑqÔq˜˜à<Ð<¨Ð<Ñ<Ô<˜åœX a­r¬zÐ:Ñ:Ô:×BÒBÀ2ÀqÑIÔIÍBÌHÐVWÐYZÐU[ÕceÔcmÐLnÑLnÔLnÑnß$šW R™[œ[ß#šV™XœXð ð
 —O’O S I¨1 IÑ.Ô.Ð.ùÝ7:¸6±{´{Ðj•”˜&­¬
Ð3Ñ3Ô3Ð3ÍÌÐQWÕ_aÔ_iÐHjÑHjÔHjˆBàð Ø_Ð_ÐV^Ð_Ñ_Ô_�Ý*¨8Ñ4Ô4��1�1�1�a�b�b�5‘	�	à�àˆhŒK×Òà&Ø ˜VØ˜a˜a˜a  1 ˜fœ:Ø     A B B œiØ (Ø"&Ø#)Ø"ð	ð 	ñô ð ñ ð ð 	ÝŒNØ”>ð cð cð cñô ð ð ×'Ò'Ñ)Ô)ˆˆ&‰	Ý ¤¨T°1Õ6KÑLÔLÐLØˆs   ®AÁAÁAr¡   c                ó6  ‡— t          | j        ¦  «                             d¦  «        }|                      ||                      ¦   «         ¦  «        \  Š}ˆfd„dD ¦   «          ‰d         }|s't          d| j        › d| j        › dt          › �¦  «        ‚t          d„ |D ¦   «         ¦  «        st          d	| j        › d
t          › �¦  «        ‚|  	                    |¦  «         d„ |D ¦   «         | _
        t          dv rt          j        d| j        › d|› �¦  «         |S )zÈLoad labels from cache or generate them from JSON file.

        Returns:
            (list[dict]): List of label dictionaries, each containing information about an image and its annotations.
        r¿   c                ó:   •— g | ]}‰                      |¦  «        ‘ŒS rŽ   rÅ   rÇ   s     €rK   rÉ   z/GroundingDataset.get_labels.<locals>.<listcomp>   s#   ø€ Ð3Ð3Ð3˜!ˆ�Š�1‰ŒÐ3Ð3Ð3rL   )r^   r¶   rR   zNo images from z
 found in r]   c              3  ó&   K  — | ]}|d          V — ŒdS )ry  NrŽ   r   s     rK   r©   z.GroundingDataset.get_labels.<locals>.<genexpr>$  s&   è è € Ð6Ð6 e�5˜”>Ð6Ð6Ð6Ð6Ð6Ð6rL   zNo annotations in ze survived filtering. Every one is iscrowd, resolves to an empty caption span or has a zero-size box. c                ó8   — g | ]}t          |d          ¦  «        ‘ŒS rË   )r4   r   s     rK   rÉ   z/GroundingDataset.get_labels.<locals>.<listcomp>*  s%   € ÐCÐCÐC°5�˜U 9Ô-Ñ.Ô.ÐCÐCÐCrL   rÀ   zLoad z from cache file )r   r�  rÌ   r½   rp   rÏ   r’  r    Úanyr§  re   r   r   rl   )rE   r¢   rx  rR   r»   s       @rK   rÓ   zGroundingDataset.get_labels  sT  ø€ õ ˜$œ.Ñ)Ô)×5Ò5°hÑ?Ô?ˆ
Ø×+Ò+¨J¸×8KÒ8KÑ8MÔ8MÑNÔN‰ˆˆqØ3Ð3Ð3Ð3Ð2Ð3Ñ3Ô3Ð3Ø�x”ˆØð 	hÝÐf°´ÐfÐfÈ4Ì=ÐfÐfÕ\dÐfÐfÑgÔgÐgÝÐ6Ð6¨vÐ6Ñ6Ô6Ñ6Ô6ð 	ÝðC T¤^ð Cð CÝ8@ðCð Cñô ð ð 	×$Ò$ VÑ,Ô,Ð,ØCÐC¸FÐCÑCÔCˆŒÝ˜Ð Ð ÝŒKÐM ¤ÐMÐMÀÐMÐMÑNÔNÐNØˆrL   NrÔ   r3   r   c                óx   •— |                       t          ¦   «                              |¦  «        | j        ¦  «        S )a  Configure augmentations for training with optional text loading.

        Args:
            hyp (dict, optional): Hyperparameters for transforms.

        Returns:
            (Compose): Composed transforms including text augmentation if applicable.
        )rø   rC   rí   rî   r{  s     €rK   rí   z!GroundingDataset.build_transforms/  s0   ø€ ð ×)Ò)­%©'¬'×*BÒ*BÀ3Ñ*GÔ*GÈÔIYÑZÔZÐZrL   c                ó$   — d„ | j         D ¦   «         S )z.Return unique category names from the dataset.c                óV   — h | ]&}|d          D ]}|D ]}|                      ¦   «         ’ŒŒŒ'S )ry  r†  )r¨   rz   r‡  rˆ  s       rK   r€  z2GroundingDataset.category_names.<locals>.<setcomp>=  s@   € Ð\Ð\Ð\˜e¸uÀW¼~Ð\Ð\°tÐW[Ð\Ð\ÐRS�—’‘	”	Ð\Ð\Ð\Ð\Ð\rL   )rR   r„   s    rK   r�  zGroundingDataset.category_names:  s   € ð ]Ð\ t¤{Ð\Ñ\Ô\Ð\rL   c                óª   — t          t          ¦  «        }| j        D ]6}|d         D ]+}|D ]&}|                     ¦   «         }||xx         dz  cc<   Œ'Œ,Œ7|S )rƒ  ry  r   )r   rŠ   rR   r~  )rE   rõ   rz   r‡  rˆ  s        rK   rõ   zGroundingDataset.category_freq?  s‚   € õ $¥CÑ(Ô(ˆØ”[ð 	*ð 	*ˆEØ˜gœð *ð *�Øð *ð *�AØŸš™	œ	�AØ! !Ð$Ð$Ô$¨Ñ)Ð$Ð$Ñ$Ð$ð*ð*ð ÐrL   )r2   r4   r�  r4   rî   rŠ   )r’  r4   rO   r�   r2  )rR   r™  rO   r£   )rN   r   rO   r¨  r6  r«   r7  )r;  r<  r=  r>  rD   r–  rp   r§  r   r€   rÓ   rí   rŒ  r�  rõ   r@  rA  s   @rK   rŽ  rŽ  q  sS  ø€ € € € € ðð ð" +3ÀRÐ\^ð Kð Kð Kð Kð Kð Kð Kð Kðð ð ð ð ð ð<ð <ð <ð <ðlð lð lð lð  )-¨Ð-=Ñ(>Ô(>ð jð jð jð jð jðXð ð ð ð.	[ð 	[ð 	[ð 	[ð 	[ð 	[ð 	[ð ð]ð ]ñ „Xð]ð ðð ñ „Xðð ð ð ð rL   rŽ  c                  ó2   — e Zd ZdZedd„¦   «         Zdd	„Zd
S )ÚYOLOConcatDatasetaã  Dataset as a concatenation of multiple datasets.

    This class is useful to assemble different existing datasets for YOLO training, ensuring they use the same collation
    function.

    Methods:
        collate_fn: Static method that collates data samples into batches using YOLODataset's collation function.

    Examples:
        >>> dataset1 = YOLODataset(...)
        >>> dataset2 = YOLODataset(...)
        >>> combined_dataset = YOLOConcatDataset([dataset1, dataset2])
    r  r¡   rO   rP   c                ó6   — t                                | ¦  «        S )r  )r.   r/  )r  s    rK   r/  zYOLOConcatDataset.collate_fnZ  s   € õ ×%Ò% eÑ,Ô,Ð,rL   rÔ   r£   c                óf   — | j         D ](}t          |d¦  «        sŒ|                     |¦  «         Œ)dS )r  r  N)ÚdatasetsÚhasattrr  )rE   rÔ   Údatasets      rK   r  zYOLOConcatDataset.close_mosaicf  sK   € ð ”}ð 	&ð 	&ˆGÝ˜7 NÑ3Ô3ð ØØ× Ò  Ñ%Ô%Ð%Ð%ð	&ð 	&rL   Nr:  r8  )r;  r<  r=  r>  r?  r/  r  rŽ   rL   rK   rç  rç  K  sR   € € € € € ðð ð ð	-ð 	-ð 	-ñ „\ð	-ð	&ð 	&ð 	&ð 	&ð 	&ð 	&rL   rç  c                  ó¨   ‡ — e Zd ZdZeZddœd.ˆ fd„Zd/d„Zd„ Zd0d„Z	d1d„Z
d2d„Zd3d„Zd4d„Zd5d„Zd6d"„Zd7ˆ fd#„Zd8ˆ fd%„	Zd9d:d*„Zd;d,„Zˆ fd-„Zˆ xZS )<ÚSemanticDatasetaõ  Dataset for semantic segmentation with PNG mask labels.

    Expects a directory structure where each image has a corresponding PNG mask file with the same stem. Pixel values in
    masks represent class IDs, with 255 as the ignore label.

    The mask directory is specified in the dataset YAML via 'masks_dir' key, and mirrors the images/ directory structure
    (e.g., images/train/ -> masks/train/).

    Attributes:
        data (dict): Dataset configuration from YAML.
        mask_files (list[str]): List of mask file paths corresponding to images.
        include_class (np.ndarray | None): Class ids to keep per pixel (None keeps all).
        masks (dict[int, np.ndarray]): Resized masks of the images in the mosaic buffer, evicted with them.
    N©r1   r1   r3   c               ó   •— |pi | _         |                      | j                              d¦  «        ¦  «        | _        |                      ¦   «         \  | _        | _        g | _        d| _        i | _	         t          ¦   «         j        |d|i|¤Ž dS )a  Initialize SemanticDataset.

        Args:
            *args (Any): Additional positional arguments for the parent class.
            data (dict): Dataset configuration dictionary.
            **kwargs (Any): Additional keyword arguments for the parent class.
        Úlabel_mappingNr1   )r1   Ú_parse_label_mappingrA   rñ  Ú_build_label_lutsÚ	label_lutÚinverse_lutÚ
mask_filesÚinclude_classr"  rC   rD   )rE   r1   rF   rG   rJ   s       €rK   rD   zSemanticDataset.__init__„  sŒ   ø€ ð �J˜BˆŒ	Ø!×6Ò6°t´y·}²}À_Ñ7UÔ7UÑVÔVˆÔØ+/×+AÒ+AÑ+CÔ+CÑ(ˆŒ˜Ô(ØˆŒØ!ˆÔØˆŒ
Ø�‰ŒÔ˜$Ð4 TÐ4¨VÐ4Ð4Ð4Ð4Ð4rL   r÷  úlist[int] | NonerO   r£   c                óN  — | j         rt          d¦  «        ‚|€dn2t          j        |t          j        ¬¦  «                             d¦  «        | _        | j        �It          | j         	                    dd¦  «        ¦  «        dk    rt          j        d¦  «         d| _        dS dS dS )	z¸Update labels to include only specified classes.

        Args:
            include_class (list[int], optional): List of classes to include. If None, all classes are included.
        zÙ'single_cls=True' is not supported for semantic segmentation: it forces a single-channel model but cannot collapse multi-class masks. Use a dataset with 'nc: 1' for binary (foreground/background) segmentation instead.Nr
  rÁ   rv   r   r   z•'classes' filtering is ignored for single-class (binary) semantic segmentation: keeping only the sole class would discard all background supervision.)rˆ   ÚNotImplementedErrorr  ÚasarrayÚint32rÂ  r÷  rŠ   r1   rA   r   ro   )rE   r÷  s     rK   Úupdate_labelszSemanticDataset.update_labels”  s¾   € ð Œ?ð 	Ý%ð@ñô ð ð
 &3Ð%:˜T˜TÅÄ
È=Õ`bÔ`hÐ@iÑ@iÔ@i×@qÒ@qÐrtÑ@uÔ@uˆÔØÔÐ)­c°$´)·-²-ÀÀaÑ2HÔ2HÑ.IÔ.IÈQÒ.NÐ.NÝŒNðKñô ð ð "&ˆDÔÐÐð *Ð)Ð.NÐ.NrL   c                óŒ  — |€i S t          |t          ¦  «        s%t          dt          |¦  «        j        › d�¦  «        ‚i }|                     ¦   «         D ]n\  }}t          |¦  «        }t          |t          ¦  «        r,|                     ¦   «         }|dk    rdnt          |¦  «        }n|€d}nt          |¦  «        }|||<   Œo|S )zNNormalize label_mapping entries from dataset YAML into integer-to-integer ids.Nz?Expected 'label_mapping' to be a dict in dataset YAML, but got r¢  Úignore_labeléÿ   )	r¹  rP   Ú	TypeErrorÚtyper;  r   rŠ   r4   r~  )rE   Úmappingr�   ÚsrcÚdsts        rK   rò  z$SemanticDataset._parse_label_mapping¨  sÓ   € àˆ?ØˆIÝ˜'¥4Ñ(Ô(ð 	yÝÐwÕ^bÐcjÑ^kÔ^kÔ^tÐwÐwÐwÑxÔxÐxàˆ
ØŸš™œð 		"ð 		"‰HˆC�Ý�c‘(”(ˆCÝ˜#�sÑ#Ô#ð Ø—i’i‘k”k�Ø  NÒ2Ð2�c�c½¸C¹¼��Ø�Ø��å˜#‘h”h�Ø!ˆJ�s‰OˆOØÐrL   útuple[np.ndarray, np.ndarray]c                ó"  — t          j        dt           j        ¬¦  «        t          j        dt           j        ¬¦  «        }}| j                             ¦   «         D ]2\  }}d|cxk    rdk     rn n|||<   d|cxk    rdk     rn Œ*|dz  ||<   Œ3||fS )zTBuild the 256-entry forward and inverse lookup tables for the dataset label mapping.é   r
  r   r   )r  ÚarangeÚuint8rñ  r   )rE   ÚforwardÚinverserÈ   rü   s        rK   ró  z!SemanticDataset._build_label_luts¼  s¥   € åœ9 Sµ´Ð9Ñ9Ô9½2¼9ÀSÕPRÔPXÐ;YÑ;YÔ;Y�ˆØÔ&×,Ò,Ñ.Ô.ð 	&ð 	&‰DˆAˆqØ�Aˆ|ˆ|Š|ˆ|˜Š|ˆ|ˆ|ˆ|ˆ|Ø�˜‘
Ø�Aˆ|ˆ|Š|ˆ|˜Š|ˆ|ˆ|ˆ|ˆ|Ø ™X�˜‘
øØ˜ÐÐrL   r�   c                óz   — t          | j        | j                             dd¦  «        d¬¦  «        | _        | j        S )zˆReturn the mask PNG paths paired with the dataset's images.

        Returns:
            (list[str]): List of mask file paths.
        Ú	masks_dirr"  rF  )Ú	label_dirÚsuffix)r$   re   r1   rA   rö  r„   s    rK   r…   zSemanticDataset.get_label_filesÆ  s7   € õ *¨$¬-À4Ä9Ç=Â=ÐQ\Ð^eÑCfÔCfÐouÐvÑvÔvˆŒØŒÐrL   r4   c                ó€   — t          j        | j        dd¬¦  «        }t          | j        | j        z   d|› �gz   ¦  «        S )z—Return a hash for semantic cache validation that also includes label_mapping changes.

        Returns:
            (str): Dataset cache hash.
        T)ú,ú:)Ú	sort_keysÚ
separatorszlabel_mapping:)rÄ  Údumpsrñ  r"   re   rö  )rE   r  s     rK   rp   zSemanticDataset.get_cache_hashÏ  sE   € õ ”*˜TÔ/¸4ÈJÐWÑWÔWˆÝ˜œ¨¬Ñ7Ð;UÈGÐ;UÐ;UÐ:VÑVÑWÔWÐWrL   rt   rŠ   rs   ru   rv   c                ó   — |› d|› d|› d�S )z6Return a one-line summary of image-mask scan counters.rŒ   z missing masks, r�   rŽ   r�   s        rK   rj   zSemanticDataset.scan_summaryØ  rT  rL   r�   c                ój   — t           t          | j        | j        t	          | j        ¦  «        ¦  «        fS )z@Return the mask verification function and its argument iterable.)r,   r’   re   rö  r   ra   r„   s    rK   rf   zSemanticDataset.verify_argsÜ  s'   € å ¥# d¤m°T´_ÅfÈTÌ[ÑFYÔFYÑ"ZÔ"ZÐZÐZrL   ry   r”   c           
     óÀ   — |\  }}}}}}}}	|rH||||t          j        g t           j        ¬¦  «        t          j        dt           j        ¬¦  «        g dddœ	nd}
|
||d||	fS )zSConvert one verify_image_mask result into a label dict and scan counter increments.r
  rW  Tr–   )	r—   Ú	mask_filer˜   Úis_1bitr™   rš   r›   r�   rž   Nr   rY  )rE   ry   r—   r  r˜   r  r{   r|   r~   r   rz   s              rK   rh   zSemanticDataset.result_to_labelà  s�   € àDJÑAˆ�˜E 7¨D°$¸¸cð ðØ"Ø&ØØ"Ý”x ­"¬*Ð5Ñ5Ô5Ýœ( 6µ´Ð<Ñ<Ô<ØØ"Ø%ð
ð 
ð 
ð ð 	ð �d˜D ! T¨3Ð.Ð.rL   rR   r¡   r¢   r   c                ó   — dS )zPSkip box and segment checks; semantic masks carry no box or segment annotations.NrŽ   r\  s      rK   r²   zSemanticDataset.verify_labelsô  r]  rL   c                ój   •— t          ¦   «                              ¦   «         }d„ |D ¦   «         | _        |S )z°Load semantic labels from cache or scan image-mask paths.

        Returns:
            (list[dict]): List of label dictionaries with mask file paths and image shapes.
        c                ó   — g | ]
}|d          ‘ŒS )r  rŽ   r§   s     rK   rÉ   z.SemanticDataset.get_labels.<locals>.<listcomp>þ  s   € Ð<Ð<Ð<¨r˜2˜kœ?Ð<Ð<Ð<rL   )rC   rÓ   rö  )rE   rR   rJ   s     €rK   rÓ   zSemanticDataset.get_labels÷  s4   ø€ õ ‘”×#Ò#Ñ%Ô%ˆØ<Ð<°VÐ<Ñ<Ô<ˆŒØˆrL   Tc                óV   •— t          ¦   «                              ||| j        ¬¦  «        S )z]Load an image for semantic segmentation, scaling the short side to imgsz when rect_mode=True.)Ú	rect_modeÚresize_short)rC   Ú
load_imagern   )rE   r-  r   rJ   s      €rK   r"  zSemanticDataset.load_image  s$   ø€ å‰wŒw×!Ò! !¨yÀtÄ|Ð!ÑTÔTÐTrL   r_  Úimage_shapeútuple[int, int] | Noneú
np.ndarrayc                óœ  — | j         |         d         }t          j        |t          j        ¦  «        }|€t	          d|› �¦  «        ‚t          | j                             dd¦  «        ¦  «        dk    r| j         |         d         r	d||dk    <   | j        r|  	                    |d	¬
¦  «        }| 
                    t          j        d	¬¦  «        S )z>Load a semantic mask and apply optional dataset label mapping.r  Nz'Semantic mask not found or unreadable: rv   r   r   r  r   F)r  )Úcopy)rR   rf  ÚimreadÚIMREAD_GRAYSCALEr·   rŠ   r1   rA   rñ  Úconvert_labelÚastyper  r
  )rE   r_  r#  r  Úmasks        rK   Ú	load_maskzSemanticDataset.load_mask  sÀ   € à”K Ô& {Ô3ˆ	ÝŒz˜)¥SÔ%9Ñ:Ô:ˆØˆ<Ý#Ð$YÈiÐ$YÐ$YÑZÔZÐZÝˆtŒy�}Š}˜T 1Ñ%Ô%Ñ&Ô&¨!Ò+Ð+°´¸EÔ0BÀ9Ô0MÐ+Ø !ˆD�˜’ÑØÔð 	;Ø×%Ò% d°EÐ%Ñ:Ô:ˆDØ�{Š{�2œ8¨%ˆ{Ñ0Ô0Ð0rL   Fc                ó†   — |r| j         n| j        }|j        t          j        k    rt          j        ||¦  «        n||         S )aZ  Convert label values using the dataset's label mapping.

        Args:
            label (np.ndarray): Segmentation label array with integer ids in 0-255.
            inverse (bool): If True, apply inverse mapping (mapped -> original). Defaults to False.

        Returns:
            (np.ndarray): New uint8 array with converted values.
        )rõ  rô  r  r  r
  rf  ÚLUT)rE   rz   r  Úluts       rK   r*  zSemanticDataset.convert_label  sA   € ð #*Ð=ˆdÔÐ¨t¬~ˆØ&+¤kµR´XÒ&=Ð&=�sŒw�u˜cÑ"Ô"Ð"À3ÀuÄ:ÐMrL   c                óT  •‡ — t          ¦   «                              |¦  «        }|d         j        dd…         \  }}‰ j                             |¦  «        }|€Ë‰                      |||f¬¦  «        }‰ j        �d|t          j        |‰ j        ¦  «         <   |j        dd…         ||fk    r#t          j
        |||ft          j        ¬¦  «        }|‰ j        v rL|‰ j        |<   t          ‰ j        ¦  «        t          ‰ j        ¦  «        k    rˆ fd„‰ j        D ¦   «         ‰ _        ||d<   |S )	aH  Get image, label and semantic mask for the given index.

        Overrides parent to include the semantic mask, served from RAM for the images Mosaic draws from the buffer.

        Args:
            index (int): Dataset index.

        Returns:
            (dict): Label dict with 'img', 'semantic_mask', and metadata.
        r  Nr;   )r#  r   rc  c                ó>   •— i | ]}|‰j         v ¯|‰j         |         “ŒS rŽ   )r"  ©r¨   r-  rE   s     €rK   rN  z7SemanticDataset.get_image_and_label.<locals>.<dictcomp>5  s*   ø€ Ð![Ð![Ð![°qÈ1ÐPTÔPZÈ?È? ! T¤Z°¤]È?È?È?rL   r  )rC   re  r˜   r"  rA   r-  r÷  r  Úisinrf  rg  rh  Úbufferrd   )rE   r_  rz   ri  rj  r,  rJ   s   `     €rK   re  z#SemanticDataset.get_image_and_label  s$  øø€ õ ‘”×+Ò+¨EÑ2Ô2ˆØ�UŒ|Ô! " 1 "Ô%‰ˆˆ1ØŒz�~Š~˜eÑ$Ô$ˆØˆ<Ø—>’> %°a¸°V�>Ñ<Ô<ˆDØÔ!Ð-Ø;>�•b”g˜d DÔ$6Ñ7Ô7Ð7Ñ8ØŒz˜"˜1˜"Œ~ ! Q Ò'Ð'Ý”z $¨¨A¨½cÔ>OÐPÑPÔP�Ø˜œÐ#Ð#Ø$(�”
˜5Ñ!Ý�t”z‘?”?¥S¨¬Ñ%5Ô%5Ò5Ð5Ø![Ð![Ð![Ð![¸D¼KÐ![Ñ![Ô![�D”JØ!%ˆˆoÑØˆrL   ©r1   r3   )r÷  rø  rO   r£   )rO   r  r1  r2  r3  r4  rn  r5  r6  )Tr«   ©r_  rŠ   r#  r$  rO   r%  )F)r;  r<  r=  r>  r   rç   rD   rý  rò  ró  r…   rp   rj   rf   rh   r²   rÓ   r"  r-  r*  re  r@  rA  s   @rK   rî  rî  r  s¦  ø€ € € € € ðð ð "€Là26ð 5ð 5ð 5ð 5ð 5ð 5ð 5ð 5ð &ð &ð &ð &ð(ð ð ð( ð  ð  ð  ðð ð ð ðXð Xð Xð Xð@ð @ð @ð @ð[ð [ð [ð [ð/ð /ð /ð /ð(_ð _ð _ð _ðð ð ð ð ð ðUð Uð Uð Uð Uð Uð
1ð 
1ð 
1ð 
1ð 
1ðNð Nð Nð Nðð ð ð ð ð ð ð ð rL   rî  c                  ó–   ‡ — e Zd ZdZddœdˆ fd„Zej        Zej        Zej        Zej	        Z	ej
        Z
ej        Zej        Zddd„Zˆ xZS )ÚPolygonSemanticDatasetaÐ  Semantic segmentation dataset that rasterizes YOLO polygon labels into masks on the fly.

    Used when the dataset YAML lacks 'masks_dir'. Pixels not covered by any polygon become a dedicated background class.
    Requires `add_polygon_background(data)` to be called first: for nc > 1 it bumps `data['nc']` to user_nc + 1 with
    background at `nc - 1`; for nc == 1 it keeps nc=1 and rasterizes a {0=bg, 1=fg} binary mask for use with
    BCEWithLogitsLoss.
    Nrï  r1   r3   c               ó(  •— |pi                       d¦  «        p$t          |pi                       di ¦  «        ¦  «        }|                      dt          t          |¦  «        dz
  d¦  «        ¦  «        | _         t          ¦   «         j        |d|i|¤Ž dS )a  Initialize PolygonSemanticDataset.

        Args:
            *args (Any): Additional positional arguments for the parent class.
            data (dict): Dataset configuration dictionary.
            **kwargs (Any): Additional keyword arguments for the parent class.
        rv   r‡   Úbg_class_idxr   r   r1   N)rA   rd   rþ   rŠ   r;  rC   rD   )rE   r1   rF   rG   rv   rJ   s        €rK   rD   zPolygonSemanticDataset.__init__C  sŽ   ø€ ð ˆj�b×Ò˜dÑ#Ô#ÐI¥s¨D¨J°B×+;Ò+;¸GÀRÑ+HÔ+HÑ'IÔ'IˆØ ŸHšH ^µS½¸R¹¼À1¹ÀaÑ5HÔ5HÑIÔIˆÔØ�‰ŒÔ˜$Ð4 TÐ4¨VÐ4Ð4Ð4Ð4Ð4rL   r_  rŠ   r#  r$  rO   r%  c                óh  ‡— |\  }}| j         |         }|                     d¦  «        }|                     d¦  «        pg }|�&t          |¦  «        dk    st          |¦  «        dk    r(t          j        ||f| j        t          j        ¬¦  «        S t          j        ||gt          j        ¬¦  «        Šˆfd„|D ¦   «         }t          ||f|d¬¦  «        \  }	}
t          j        ||f| j        t          j        ¬¦  «        }|	dk    }t          | j                             d	d¦  «        ¦  «        dk    rd||<   n{t          j        |¦  «                             d
¦  «                             t          j        ¦  «        |
         }||	|         dz
                                t          j        ¦  «        ||<   |S )zZRasterize this image's polygons into a (H, W) uint8 semantic mask, bg = self.bg_class_idx.r™   r›   Nr   r
  c                ó~   •— g | ]9}t          j        |t           j        ¬ ¦  «                             dd¦  «        ‰z  ‘Œ:S )r
  rÁ   r;   )r  rû  r  rÂ  )r¨   r  Úscales     €rK   rÉ   z4PolygonSemanticDataset.load_mask.<locals>.<listcomp>e  s?   ø€ ÐZÐZÐZÈA•”˜A¥R¤ZÐ0Ñ0Ô0×8Ò8¸¸QÑ?Ô?À%ÑGÐZÐZÐZrL   r   )Údownsample_ratiorv   rÁ   )rR   rA   rd   r  Úfullr;  r
  rZ  r  r'   rŠ   r1   rû  rÂ  r+  rü  )rE   r_  r#  ri  rj  rz   r™   r›   ÚpolysÚinstÚ
sorted_idxÚoutÚfgÚcls_arrr>  s                 @rK   r-  z PolygonSemanticDataset.load_maskZ  s‡  ø€ à‰ˆˆ1Ø”˜EÔ"ˆØ�iŠi˜ÑÔˆØ—9’9˜ZÑ(Ô(Ð.¨BˆØˆ;�#˜c™(œ( aš-˜-­3¨x©=¬=¸AÒ+=Ð+=Ý”7˜A˜q˜6 4Ô#4½B¼HÐEÑEÔEÐEõ ”˜!˜Q˜¥r¤zÐ2Ñ2Ô2ˆØZÐZÐZÐZÐQYÐZÑZÔZˆå1°1°a°&¸%ÐRSÐTÑTÔTÑˆˆjÝŒg�q˜!�f˜dÔ/µr´xÐ@Ñ@Ô@ˆØ�AŠXˆÝˆtŒy�}Š}˜T 1Ñ%Ô%Ñ&Ô&¨!Ò+Ð+ØˆC�‰GˆGå”j ‘o”o×-Ò-¨bÑ1Ô1×8Ò8½¼ÑBÔBÀ:ÔNˆGØ˜d 2œh¨™lÔ+×2Ò2µ2´8Ñ<Ô<ˆC�‰GØˆ
rL   r6  r«   r7  )r;  r<  r=  r>  rD   r.   rÓ   r…   rp   rj   rf   rh   r²   r-  r@  rA  s   @rK   r9  r9  :  s©   ø€ € € € € ðð ð 37ð 
5ð 
5ð 
5ð 
5ð 
5ð 
5ð 
5ð 
5ð Ô'€JØ!Ô1€OØ Ô/€NØÔ+€LØÔ)€KØ!Ô1€OØÔ-€Mðð ð ð ð ð ð ð ð rL   r9  c                  ó<   — e Zd ZdZddd„Zdd„Zdd„Zdd„Zdd„ZdS )ÚClassificationDatasetaÅ  Dataset class for image classification tasks wrapping torchvision ImageFolder functionality.

    This class offers functionalities like image augmentation, caching, and verification. It's designed to efficiently
    handle large datasets for training deep learning models, with optional image transformations and caching mechanisms
    to speed up training.

    Attributes:
        cache_ram (bool): Indicates if caching in RAM is enabled.
        cache_disk (bool): Indicates if caching on disk is enabled.
        samples (list): A list of lists, each containing the path to an image, its class index, path to its .npy cache
            file (if caching on disk), and optionally the loaded image array (if caching in RAM).
        torch_transforms (callable): PyTorch transforms to be applied to the images.
        root (str): Root directory of the dataset.
        prefix (str): Prefix for logging and cache filenames.

    Methods:
        __getitem__: Return transformed image and class index for the given sample index.
        __len__: Return the total number of samples in the dataset.
        verify_images: Verify all images in dataset.
        cache_images: Decode images into one contiguous RAM cache.
    Fr�  NÚrootr4   rn   rè   ra   r‡   údict[int, str] | Nonec                óÚ  ‡ ‡‡‡‡— ddl }t          r"|j                             |d¬¦  «        ‰ _        n |j                             |¬¦  «        ‰ _        t          |¦  «        j        dz                       ¦   «         }‰ j        j        ‰ _        ‰ j        j	        ‰ _	        |rdnt          |j        |p|rdnd	¦  «        }t          |t          ¦  «        r|nBt          t          |dk    ¦  «        t          t!          ‰ j        ¦  «        |z  ¦  «        ¦  «        }	|	t!          ‰ j        ¦  «        k     r>ˆ fd
„t#          j        dt!          ‰ j        ¦  «        dz
  |	t          ¬¦  «        D ¦   «         n‰ j        ‰ _        |rt'          |› d�¦  «        nd‰ _        |j        du p)t-          |j        ¦  «                             ¦   «         dk    ‰ _        t-          |j        ¦  «                             ¦   «         dk    ‰ _        ‰                      ¦   «         ‰ _        ‰ j        j        Š|j        rt:                               ‰d¦  «        Šn}|rd„ ‰D ¦   «         Šnn‰rSt?          ‰¦  «                              ‰ !                    ¦   «         ¦  «        sd„ ‰ "                    ¦   «         D ¦   «         Šnd„ tG          ‰¦  «        D ¦   «         Š‰rˆˆfd„‰D ¦   «         nt?          ¦   «         Št!          ‰ j        ¦  «        }
ˆˆˆfd„‰ j        D ¦   «         ‰ _        ‰rCtI          j%        ‰ j        › d|
t!          ‰ j        ¦  «        z
  › dtM          ‰¦  «        › �¦  «         d„ ‰ j        D ¦   «         ‰ _        ‰ j        r‰  '                    ¦   «          d|j(        z
  df}|r@tS          |j*        ||j+        |j,        |j-        |j.        |j/        |j0        |j1        ¬¦	  «	        nte          |j*        ¬¦  «        ‰ _3        dS )a>  Initialize YOLO classification dataset with root directory, arguments, augmentations, and cache settings.

        Args:
            root (str): Path to the dataset directory where images are stored in a class-specific folder structure.
            args (Namespace): Configuration containing dataset-related settings such as image size, augmentation
                parameters, and cache settings.
            augment (bool, optional): Whether to apply augmentations to the dataset.
            prefix (str, optional): Prefix for logging and cache filenames, aiding in dataset identification.
            names (dict[int, str], optional): Model class names; class folders are aligned to this order by name and
                folders the model lacks are dropped, since each split's ImageFolder scan is indexed on its own.
        r   NT)rI  Úallow_empty)rI  z.ndjson.yamlr”  ÚtrainÚvalc                ó*   •— g | ]}‰j         |         ‘ŒS rŽ   )Úsamplesr3  s     €rK   rÉ   z2ClassificationDataset.__init__.<locals>.<listcomp>¤  s   ø€ Ð^Ð^Ð^ ˆTŒ\˜!Œ_Ð^Ð^Ð^rL   r   r
  z: r�  ÚramÚdiskc                ó.   — i | ]}|t          |¦  «        “ŒS rŽ   )rŠ   r»  s     rK   rN  z2ClassificationDataset.__init__.<locals>.<dictcomp>°  s    € Ð0Ð0Ð0 1�Q�˜A™œÐ0Ð0Ð0rL   c                ó   — i | ]\  }}||“Œ	S rŽ   rŽ   )r¨   r-  rÐ   s      rK   rN  z2ClassificationDataset.__init__.<locals>.<dictcomp>²  s   € Ð4Ð4Ð4™d˜a �Q˜Ð4Ð4Ð4rL   c                ó   — i | ]\  }}||“Œ	S rŽ   rŽ   )r¨   r-  r‹  s      rK   rN  z2ClassificationDataset.__init__.<locals>.<dictcomp>´  s   € Ð9Ð9Ð9™d˜a �Q˜Ð9Ð9Ð9rL   c                ó|   •— h | ]8}‰                      |t          ‰¦  «        ¦  «        t          ‰¦  «        k    ¯6|’Œ9S rŽ   )rA   rd   )r¨   r‹  r_  r‡   s     €€rK   r€  z1ClassificationDataset.__init__.<locals>.<setcomp>µ  s=   ø€ ÐNÐNÐN�q u§y¢y°µC¸±J´JÑ'?Ô'?Å3ÀuÁ:Ä:Ò'MÐ'M�Ð'MÐ'MÐ'MrL   c                óJ   •— g | ]\  }}‰|         ‰v¯|‰‰|                  f‘Œ S rŽ   rŽ   )r¨   rM  rˆ  ÚclassesÚextrar_  s      €€€rK   rÉ   z2ClassificationDataset.__init__.<locals>.<listcomp>·  s;   ø€ ÐcÐcÐc±4°1°aÈ7ÐSTÌ:Ð]bÐKbÐKb˜˜E '¨!¤*Ô-Ð.ÐKbÐKbÐKbrL   z	Skipping z' samples from classes the model lacks: c                ó„   — g | ]=}g t          |¦  «        ¢t          |d          ¦  «                             d¦  «        ‘d‘‘Œ>S )r   rG  N)r“   r   rÌ   r­   s     rK   rÉ   z2ClassificationDataset.__init__.<locals>.<listcomp>¼  sF   € Ð_Ð_Ð_ÈQÐH�$˜q™'œ'ÐH¥4¨¨!¬¡:¤:×#9Ò#9¸&Ñ#AÔ#AÐHÀ4ÐHÐ_Ð_Ð_rL   )	Úsizer>  ÚhflipÚvflipÚerasingÚauto_augmentÚhsv_hÚhsv_sÚhsv_v)r[  )4Útorchvisionr   rê  ÚImageFolderÚbaser   rb   rI  rP  rI  r#   r•  r¹  rŠ   rþ   Úroundrd   r  Úlinspacer   ra   r»   r4   rÆ  Ú	cache_ramÚ
cache_diskÚverify_imagesrX  rˆ   rP   rŠ  rÎ   Ú
isdisjointrÿ   r   r$  r   ro   rÍ   Úcache_imagesr>  r   ræ   ÚfliplrÚflipudr^  r_  r`  ra  rb  r   Útorch_transforms)rE   rI  rF   rn   ra   r‡   rc  Ú	is_ndjsonr•  r¦  rÐ   r>  rX  rY  r_  s   `    `      @@@rK   rD   zClassificationDataset.__init__‰  sü  øøøøø€ ð 	ÐÐÐõ ð 	DØ#Ô,×8Ò8¸dÐPTÐ8ÑUÔUˆDŒIˆIà#Ô,×8Ò8¸dÐ8ÑCÔCˆDŒIÝ˜$‘Z”ZÔ&¨Ñ7×@Ò@ÑBÔBˆ	Ø”yÔ(ˆŒØ”I”NˆŒ	ð $Ðu�3�3Õ);¸D¼MÈ6ÐKtÐahÐVsÐV]ÐV]ÐnsÑ)uÔ)uˆÝ& xµÑ5Ô5Ðv��½3½sÀ8ÈaÂ<Ñ?PÔ?PÕRWÕX[Ð\`Ô\hÑXiÔXiÐltÑXtÑRuÔRuÑ;vÔ;vˆð •s˜4œ<Ñ(Ô(Ò(Ð(ð _Ð^Ð^Ð^¥b¤k°!µS¸¼Ñ5FÔ5FÈÑ5JÈEÕY\Ð&]Ñ&]Ô&]Ð^Ñ^Ô^Ð^à”ð 	Œð
 28Ð?•h &˜}˜}˜}Ñ-Ô-Ð-¸RˆŒØœ tÐ+ÐO­s°4´:©¬×/DÒ/DÑ/FÔ/FÈ%Ò/OˆŒÝ˜dœj™/œ/×/Ò/Ñ1Ô1°VÒ;ˆŒØ×)Ò)Ñ+Ô+ˆŒØ”)Ô#ˆØŒ?ð 	:Ý—M’M '¨1Ñ-Ô-ˆEˆEØð 	:Ø0Ð0¨Ð0Ñ0Ô0ˆEˆEØð 	:�3˜w™<œ<×2Ò2°5·<²<±>´>ÑBÔBð 	:Ø4Ð4 e§k¢k¡m¤mÐ4Ñ4Ô4ˆEˆEà9Ð9¥i°Ñ&8Ô&8Ð9Ñ9Ô9ˆEØRWÐbÐNÐNÐNÐNÐN˜GÐNÑNÔNÐNÕ]`Ñ]bÔ]bˆÝ�”ÑÔˆØcÐcÐcÐcÐcÐc¸4¼<ÐcÑcÔcˆŒØð 	ÝŒNØ”;ÐvÐv¨­S°´Ñ->Ô->Ñ)>ÐvÐvÕgmÐnsÑgtÔgtÐvÐvñô ð ð `Ð_ÐRVÔR^Ð_Ñ_Ô_ˆŒØŒ>ð 	 Ø×ÒÑÔÐØ�t”zÑ! 3Ð'ˆð ð6Õ"Ø”ZØØ”kØ”kØœØ!Ô.Ø”jØ”jØ”jð
ñ 
ô 
ð 
õ %¨$¬*Ð5Ñ5Ô5ð 	ÔÐÐrL   r-  rŠ   rO   rP   c                óä  — | j         |         \  }}}}| j        r| j        |         }n| j        rd|                     ¦   «         s;t          j        |                     ¦   «         t          j	        |¦  «        d¬¦  «         t          j
        |¦  «        }nt          j	        |¦  «        }t          j        t          j        |t          j        ¦  «        ¦  «        }|                      |¦  «        }||dœS )zíReturn transformed image and class index for the given sample index.

        Args:
            i (int): Index of the sample to retrieve.

        Returns:
            (dict): Dictionary containing the image and its class index.
        F)Úallow_pickle)r  r™   )rP  rh  Ú	img_cacheri  r¼   r  ÚsaveÚas_posixrf  r(  rÅ  r	   Ú	fromarrayÚcvtColorÚCOLOR_BGR2RGBro  )rE   r-  rM  r¿  ÚfnÚimÚsamples          rK   Ú__getitem__z!ClassificationDataset.__getitem__Ð  sÌ   € ð ”| A”‰ˆˆ1ˆb�"ØŒ>ð 	Ø” Ô"ˆBˆBØŒ_ð 	Ø—9’9‘;”;ð JÝ”˜Ÿš™œ¥s¤z°!¡}¤}À5ÐIÑIÔIÐIÝ”˜‘”ˆBˆBå”˜A‘”ˆBåŒ_�Sœ\¨"­cÔ.?Ñ@Ô@ÑAÔAˆØ×&Ò& rÑ*Ô*ˆØ aÐ(Ð(Ð(rL   c                ó*   — t          | j        ¦  «        S )z2Return the total number of samples in the dataset.)rd   rP  r„   s    rK   Ú__len__zClassificationDataset.__len__ç  s   € å�4”<Ñ Ô Ð rL   r£   c           
     óJ  — t          t          ¦  «        5 }t          t          |                     d„ | j        ¦  «        t          | j        ¦  «        | j        › d�t          dk    ¬¦  «        ¦  «        }ddd¦  «         n# 1 swxY w Y   t          j
        |¦  «        | _        dS )aŸ  Decode all images once into a single contiguous uint8 buffer before DataLoader workers fork.

        A Python list of per-image arrays is duplicated into every forked worker by copy-on-write refcounting
        (https://github.com/ultralytics/ultralytics/issues/9824); one shared buffer is read-only across
        workers instead, so RAM stays flat. Original image sizes are preserved for the transforms.
        c                ó6   — t          j        | d         ¦  «        S )Nr   )rf  r(  )r  s    rK   ú<lambda>z4ClassificationDataset.cache_images.<locals>.<lambda>õ  s   € ­¬
°1°Q´4Ñ(8Ô(8€ rL   zCaching imagesr   )rZ   rY   ÚdisableN)r   r   r“   r   rg   rP  rd   ra   r   r   Ú_ImageCachers  )rE   rw   Úimss      rK   rl  z"ClassificationDataset.cache_imagesë  sÖ   € õ �Ñ$Ô$ð 	¨ÝÝØ—I’IÐ8Ð8¸$¼,ÑGÔGÝ˜dœlÑ+Ô+Ø œKÐ7Ð7Ð7Ý&¨šNð	ñ ô ñô ˆCð	ð 	ð 	ñ 	ô 	ð 	ð 	ð 	ð 	ð 	ð 	øøøð 	ð 	ð 	ð 	õ %Ô0°Ñ5Ô5ˆŒˆˆs   •AA?Á?BÂBúlist[tuple]c                óâ  — | j         › d| j        › d�}t          | j        ¦  «                             d¦  «        }	 t	          d„ | j        dd…         D ¦   «         | j         ¬¦  «         t          |¦  «        }|d         t          k    sJ ‚|d	         t          d
„ | j        D ¦   «         | j	        j
        z   ¦  «        k    sJ ‚|                     d¦  «        \  }}}}t          dv rS|› d|› d|› d�}t          d|||¬¦  «         |d         r-t          j        d                     |d         ¦  «        ¦  «         |S # t"          t$          t&          t(          f$ �r— ddg g i f\  }}}	}}
t+          t,          ¦  «        5 }|                     t0          t3          | j        t5          | j         ¦  «        ¦  «        ¬¦  «        }t          ||t7          | j        ¦  «        ¬¦  «        }|D ]O\  }}}}|r|                     |¦  «         |r|	                     |¦  «         ||z  }||z  }|› d|› d|› d�|_        ŒP|                     ¦   «          ddd¦  «         n# 1 swxY w Y   |	r't          j        d                     |	¦  «        ¦  «         t          d„ | j        D ¦   «         | j	        j
        z   ¦  «        |
d	<   ||t7          |¦  «        |f|
d<   |	|
d<   t?          | j         ||
t          ¦  «         |cY S w xY w)z}Verify all images in dataset.

        Returns:
            (list[tuple]): List of valid samples after verification.
        rS   rT   r¿   c                ó   — g | ]\  }}|‘ŒS rŽ   rŽ   )r¨   Úfilerx  s      rK   rÉ   z7ClassificationDataset.verify_images.<locals>.<listcomp>  s   € ÐFÐFÐF©	¨¨q˜tÐFÐFÐFrL   NrÀ  )ra   r¶   r^   c                ó   — g | ]
}|d          ‘ŒS ©r   rŽ   r­   s     rK   rÉ   z7ClassificationDataset.verify_images.<locals>.<listcomp>
  s   € Ð-IÐ-IÐ-I°q¨a°¬dÐ-IÐ-IÐ-IrL   r_   rÀ   r[   rŒ   r�   rÂ   r`   r\   r   rU   rX   c                ó   — g | ]
}|d          ‘ŒS rŠ  rŽ   r­   s     rK   rÉ   z7ClassificationDataset.verify_images.<locals>.<listcomp>%  s   € Ð!=Ð!=Ð!=¨1 ! A¤$Ð!=Ð!=Ð!=rL   ) ra   rI  r   rÌ   r!   rP  r%   rq   r"   re  rX  rÆ   r   r   r   rl   rm   r·   r¸   r¹   rº   r   r   rg   r)   r’   r   rd   ri   rY   rk   r(   )rE   rY   rN   r»   rt   rv   rÐ   rP  rÑ   r`   rr   rw   r_   rx   r{  r|   r~   r   s                     rK   rj  z#ClassificationDataset.verify_imagesý  sb  € ð ”+Ð6Ð6¨¬	Ð6Ð6Ð6ˆÝ�D”I‰Œ×*Ò*¨8Ñ4Ô4ˆð#	ÝÐFÐF°T´\À"À1À"Ô5EÐFÑFÔFÈtÌ{Ð[Ñ[Ô[Ð[Ý+¨DÑ1Ô1ˆEØ˜Ô#Õ'<Ò<Ð<Ð<Ð<Ø˜”=¥HÐ-IÐ-I¸D¼LÐ-IÑ-IÔ-IÈDÌIÔL]Ñ-]Ñ$^Ô$^Ò^Ð^Ð^Ð^Ø!&§¢¨9Ñ!5Ô!5ÑˆB��A�wÝ˜WÐ$Ð$ØÐ7Ð7˜bÐ7Ð7¨2Ð7Ð7Ð7�Ý�T ¨°AÐ6Ñ6Ô6Ð6Ø˜”=ð :Ý”K §	¢	¨%°¬-Ñ 8Ô 8Ñ9Ô9Ð9ØˆNøõ "¥>µ>ÕCVÐWð 	ñ 	ð 	à'(¨!¨R°°RÐ'7Ñ$ˆB��D˜' 1Ý�KÑ(Ô(ð ¨DØŸ)š)­ÅÀDÄLÕRXÐY]ÔYdÑReÔReÑ@fÔ@f˜)ÑgÔg�Ý˜G¨$µc¸$¼,Ñ6GÔ6GÐHÑHÔH�Ø/3ð Dð DÑ+�F˜D $¨Øð /ØŸš vÑ.Ô.Ð.Øð )ØŸš CÑ(Ô(Ð(Ø˜$‘J�BØ˜$‘J�BØ#'Ð CÐ C¨"Ð CÐ C°rÐ CÐ CÐ C�D”I�IØ—
’
‘”�ðð ð ñ ô ð ð ð ð ð ð øøøð ð ð ð ð ð -Ý”˜DŸIšI d™OœOÑ,Ô,Ð,Ý Ð!=Ð!=°´Ð!=Ñ!=Ô!=ÀÄ	Ô@QÑ!QÑRÔRˆAˆf‰IØ˜r¥3 w¡<¤<°Ð8ˆAˆi‰LØˆAˆf‰IÝ# D¤K°°qÕ:OÑPÔPÐPØˆNˆNˆNð+	øøøs8   »C=D9 Ä9?K.Å8CIÉK.ÉI	ÉK.ÉI	ÉBK.Ë-K.)Fr�  N)rI  r4   rn   rè   ra   r4   r‡   rJ  )r-  rŠ   rO   rP   )rO   rŠ   )rO   r£   )rO   r…  )	r;  r<  r=  r>  rD   r|  r~  rl  rj  rŽ   rL   rK   rH  rH  r  s�   € € € € € ðð ð,E
ð E
ð E
ð E
ð E
ðN)ð )ð )ð )ð.!ð !ð !ð !ð6ð 6ð 6ð 6ð$,ð ,ð ,ð ,ð ,ð ,rL   rH  )GÚ
__future__r   rÄ  Úcollectionsr   Ú	itertoolsr   Úmultiprocessing.poolr   Úpathlibr   Útypingr   rf  Únumpyr  r%  ÚPILr	   Útorch.utils.datar
   Úultralytics.utilsr   r   r   r   r   Úultralytics.utils.instancer   Úultralytics.utils.opsr   r   Úultralytics.utils.torch_utilsr   rn   r   r   r   r   r   r   r   r   r   re  r   Ú	converterr   r'  r    r!   r"   r#   r$   r%   r&   r'   r(   r)   r*   r+   r,   rq   r.   rC  rp  rŽ  rç  rî  r9  rH  rŽ   rL   rK   ú<module>rš     sñ  ðð #Ð "Ð "Ð "Ð "Ð "à €€€Ø #Ð #Ð #Ð #Ð #Ð #Ø Ð Ð Ð Ð Ð Ø +Ð +Ð +Ð +Ð +Ð +Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð à 
€
€
€
Ø Ð Ð Ð Ø €€€Ø Ð Ð Ð Ð Ð Ø *Ð *Ð *Ð *Ð *Ð *à MÐ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MØ 0Ð 0Ð 0Ð 0Ð 0Ð 0Ø CÐ CÐ CÐ CÐ CÐ CÐ CÐ CØ :Ð :Ð :Ð :Ð :Ð :ð
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð Ð Ð Ð Ð Ð Ø *Ð *Ð *Ð *Ð *Ð *ðð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð"  Ð ð@ð @ð @ð @ð @�+ñ @ô @ð @ðFdð dð dð dð d�;ñ dô dð dðNL^ð L^ð L^ð L^ð L^˜Kñ L^ô L^ð L^ð^Wð Wð Wð Wð W�{ñ Wô Wð Wðt$&ð $&ð $&ð $&ð $&˜ñ $&ô $&ð $&ðNEð Eð Eð Eð E�kñ Eô Eð EðP5ð 5ð 5ð 5ð 5˜_¨kñ 5ô 5ð 5ðpwð wð wð wð wñ wô wð wð wð wrL   