§
    l¬¸j$  ã                  óø   — d dl mZ d dlmZ d dlmZ d dlZd dlmZ	 d dl
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 d dlmZ d dlmZ d dlmZmZ d dlmZm Z   G d„ de¦  «        Z! G d„ de!e¦  «        Z"dS )é    )Úannotations)ÚPath)ÚAnyN)Ú
functional)ÚYOLOConcatDatasetÚbuild_dataloaderÚbuild_yolo_dataset)ÚLoadVisualPrompt)Úcheck_det_datasetÚget_split_fraction)ÚDetectionValidator)ÚSegmentationValidator)ÚYOLOEDetect)Ú
YOLOEModel)ÚLOGGERÚTQDM)Úselect_deviceÚsmart_inference_modec                  ó€   ‡ — e Zd ZdZ e¦   «         dd„¦   «         Zdd„Z ed¦  «        	 	 	 	 ddˆ fd„¦   «         Zˆ xZS )ÚYOLOEDetectValidatoraö  A validator class for YOLOE detection models that handles both text and visual prompt embeddings.

    This class extends DetectionValidator to provide specialized validation functionality for YOLOE models. It supports
    validation using either text prompts or visual prompt embeddings extracted from training samples, enabling flexible
    evaluation strategies for prompt-based object detection.

    Attributes:
        device (torch.device): The device on which validation is performed.
        args (namespace): Configuration arguments for validation.
        dataloader (DataLoader): DataLoader for validation data.

    Methods:
        get_visual_pe: Extract visual prompt embeddings from training samples.
        preprocess: Preprocess batch data ensuring visuals are on the same device as images.
        get_vpe_dataloader: Create a dataloader for LVIS training visual prompt samples.
        get_model: Prepare a validation model with text or visual prompt embeddings.

    Examples:
        Validate with text prompts
        >>> validator = YOLOEDetectValidator()
        >>> stats = validator(model=model, load_vp=False)

        Validate with visual prompts
        >>> stats = validator(model=model, refer_data="path/to/data.yaml", load_vp=True)
    Ú
dataloaderútorch.utils.data.DataLoaderÚmodelr   Úreturnútorch.Tensorc           	     ó¶  — t          |t          ¦  «        sJ ‚d„ t          |j        j        d                              ¦   «         ¦  «        D ¦   «         }t          j        t          |¦  «        |j	        d         j
        | j        ¬¦  «        }t          j        t          |¦  «        ¦  «        }d}|D ]t}|d                              d¦  «                             t          j        ¦  «                             ¦   «         }t          j        |t          |¦  «        ¬¦  «        }	||	z  }Œu|                     | j        ¦  «        }t#          |t          |¦  «        |¬¦  «        }
|
D �]2}|                      |¦  «        }|                     |d	         |d
         ¬¦  «        }|d         }t)          |j        d         ¦  «        D ]Ô}|d         ||k                                  d¦  «                             t          j        ¦  «                             d¬¦  «        }t          j        |j        d         | j        ¬¦  «        dz  }||d|j        d         …<   |D ]>}||xx         ||         ||k                                  d¦  «        ||         z  z  cc<   Œ?ŒÕ�Œ4t1          j        ||dk             dd¬¦  «        ||dk    <   d||dk    <   |                     d¦  «        S )az  Extract visual prompt embeddings from training samples.

        This method processes a dataloader to compute visual prompt embeddings for each class using a YOLOE model. It
        normalizes the embeddings and handles cases where no samples exist for a class by setting their embeddings to
        zero.

        Args:
            dataloader (torch.utils.data.DataLoader): The dataloader providing training samples.
            model (YOLOEModel): The YOLOE model from which to extract visual prompt embeddings.

        Returns:
            (torch.Tensor): Visual prompt embeddings with shape (1, num_classes, embed_dim).
        c                óF   — g | ]}|                      d d¦  «        d         ‘ŒS ©ú/é   r   ©Úsplit©Ú.0Únames     úq/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/ultralytics/models/yolo/yoloe/val.pyú
<listcomp>z6YOLOEDetectValidator.get_visual_pe.<locals>.<listcomp>A   s+   € ÐcÐcÐc¨4�—’˜C Ñ#Ô# AÔ&ÐcÐcÐcó    Únameséÿÿÿÿ©Údevicez)Get visual prompt embeddings from samplesÚcls)Ú	minlength)ÚtotalÚdescÚimgÚvisuals)ÚvisualÚ	batch_idxr   T)Úsortedr    Né   )ÚdimÚp)Ú
isinstancer   ÚlistÚdatasetÚdataÚvaluesÚtorchÚzerosÚlenr   Úembedr,   ÚsqueezeÚtoÚintÚuniqueÚbincountr   Ú
preprocessÚget_visual_peÚrangeÚshapeÚonesÚsumÚFÚ	normalizeÚ	unsqueeze)Úselfr   r   r)   Ú	visual_peÚcls_visual_numr0   Úbatchr-   ÚcountÚpbarÚpredsr4   ÚiÚpad_clsÚcs                   r&   rH   z"YOLOEDetectValidator.get_visual_pe1   s¹  € õ ˜%¥Ñ,Ô,Ð,Ð,Ð,ØcÐcµ4¸
Ô8JÔ8OÐPWÔ8X×8_Ò8_Ñ8aÔ8aÑ3bÔ3bÐcÑcÔcˆÝ”K¥ E¡
¤
¨E¬K¸¬OÔ,AÈ$Ì+ÐVÑVÔVˆ	Ýœ¥S¨¡Z¤ZÑ0Ô0ˆà:ˆð  ð 	$ð 	$ˆEØ˜”,×&Ò& rÑ*Ô*×-Ò-­e¬iÑ8Ô8×?Ò?ÑAÔAˆCÝ”N 3µ#°e±*´*Ð=Ñ=Ô=ˆEØ˜eÑ#ˆNˆNà'×*Ò*¨4¬;Ñ7Ô7ˆõ �J¥c¨*¡o¤o¸DÐAÑAÔAˆØð 
	Vñ 
	VˆEØ—O’O EÑ*Ô*ˆEØ×'Ò'¨¨e¬¸UÀ9Ô=MÐ'ÑNÔNˆEà˜kÔ*ˆIÝ˜5œ; qœ>Ñ*Ô*ð Vð V�Ø˜E”l 9°¢>Ô2×:Ò:¸2Ñ>Ô>×AÒAÅ%Ä)ÑLÔL×SÒSÐ[_ÐSÑ`Ô`�Ýœ* U¤[°¤^¸D¼KÐHÑHÔHÈ2ÑM�Ø*-�˜˜#œ) Aœ,˜Ñ'Øð Vð V�AØ˜a�L�L”L E¨!¤H¨W¸ª\Ô$:×$>Ò$>¸qÑ$AÔ$AÀNÐSTÔDUÑ$UÑU�L�L‘L�LðVñ	Võ *+¬°Y¸~ÐQRÒ?RÔ5SÐY[Ð_`Ð)aÑ)aÔ)aˆ	�. AÒ%Ñ&Ø)*ˆ	�. AÒ%Ñ&Ø×"Ò" 1Ñ%Ô%Ð%r(   r<   údict[str, Any]c                ó*  — t          | j        |                     | j        j        |                     d¦  «        ¦  «        | j        j        |ddt          | j        j        | j        j        pd¦  «        ¬¦  «        }t          |t          ¦  «        r1|j	        D ](}|j
                             t          ¦   «         ¦  «         Œ)n&|j
                             t          ¦   «         ¦  «         t          || j        j        | j        j        dd| j        ¬¦  «        S )a  Create a dataloader for LVIS training visual prompt samples.

        This method prepares a dataloader for visual prompt embeddings (VPE) using the specified dataset. It applies
        necessary transformations including LoadVisualPrompt and configurations to the dataset for validation purposes.

        Args:
            data (dict): Dataset configuration dictionary containing paths and settings.

        Returns:
            (torch.utils.data.DataLoader): The dataloader for visual prompt samples.
        ÚvalF)ÚmodeÚrectÚfractionr*   )ÚshuffleÚrankr,   )r	   ÚargsÚgetr"   rS   r   r_   r9   r   ÚdatasetsÚ
transformsÚappendr
   r   Úworkersr,   )rP   r<   r;   Úds       r&   Úget_vpe_dataloaderz'YOLOEDetectValidator.get_vpe_dataloaderb   s  € õ %ØŒIØ�HŠH�T”Y”_ d§h¢h¨u¡o¤oÑ6Ô6ØŒIŒOØØØÝ'¨¬	Ô(:¸D¼I¼OÐ<TÈuÑUÔUð
ñ 
ô 
ˆõ �gÕ0Ñ1Ô1ð 	:ØÔ%ð 8ð 8�Ø”×#Ò#Õ$4Ñ$6Ô$6Ñ7Ô7Ð7Ð7ð8ð Ô×%Ò%Õ&6Ñ&8Ô&8Ñ9Ô9Ð9ÝØØŒIŒOØŒIÔØØØ”;ð
ñ 
ô 
ð 	
r(   FNúYOLOEModel | str | NoneÚtrainerú
Any | NoneÚ
refer_dataú
str | NoneÚload_vpÚboolc                ód  •— t          ¦   «                              ||¦  «        }|�â|j        | _        d„ t          | j        j        j        d                              ¦   «         ¦  «        D ¦   «         }|rSt          j	        d¦  «         d| j
        _        |                      | j        |¦  «        }|                     ||¦  «         �nit          j	        d¦  «         |                     |¦  «        }|                     ||¦  «         �n(|�|s
J d¦   «         ‚t          | j
        j        d¬¦  «        | _        t!          |t"          t$          f¦  «        rd	d
lm}  ||| j        ¬¦  «        \  }}	|                     ¦   «                              | j        ¦  «         t/          |p| j
        j        ¦  «        }
d„ t          |
d                              ¦   «         ¦  «        D ¦   «         }|�kt/          | j
        j        ¦  «        }d„ t          |d                              ¦   «         ¦  «        D ¦   «         }||k    rt          j        d|› d|› d�¦  «         |rbt          j	        d¦  «         d| j
        _        |                      |
¦  «        }|                      ||¦  «        }|                     ||¦  «         nzt!          |j        d         t6          ¦  «        rt9          |j        d         d¦  «        s?t          j	        d¦  «         |                     |¦  «        }|                     ||¦  «         |S )a.  Prepare text, visual, or prompt-free models before validation inference setup.

        Args:
            trainer (object, optional): Trainer object containing the model and device.
            model (YOLOEModel | str, optional): Model to validate. Required if trainer is not provided.
            refer_data (str, optional): Path to reference data for visual prompts.
            load_vp (bool): Whether to load visual prompts. If False, text prompts are used.

        Returns:
            (YOLOEModel): Model with prompts prepared for validation.
        Nc                óF   — g | ]}|                      d d¦  «        d         ‘ŒS r   r!   r#   s     r&   r'   z2YOLOEDetectValidator.get_model.<locals>.<listcomp>›   s+   € ÐlÐlÐl¨t�T—Z’Z  QÑ'Ô'¨Ô*ÐlÐlÐlr(   r)   z!Validate using the visual prompt.zValidate using the text prompt.z5Refer data is only used for visual prompt validation.F)Úverboser   )Úload_checkpointr+   c                óF   — g | ]}|                      d d¦  «        d         ‘ŒS r   r!   r#   s     r&   r'   z2YOLOEDetectValidator.get_model.<locals>.<listcomp>²   s+   € ÐTÐTÐT¨t�T—Z’Z  QÑ'Ô'¨Ô*ÐTÐTÐTr(   c                óF   — g | ]}|                      d d¦  «        d         ‘ŒS r   r!   r#   s     r&   r'   z2YOLOEDetectValidator.get_model.<locals>.<listcomp>¶   s+   € ÐbÐbÐb¸˜dŸjšj¨¨aÑ0Ô0°Ô3ÐbÐbÐbr(   zClass names from refer data z! do not match evaluation dataset z0. This may lead to incorrect validation results.r*   Úlrpc)ÚsuperÚ	get_modelr,   r:   r   r;   r<   r=   r   Úinforb   ÚquantizerH   Úset_classesÚget_text_per   r9   Ústrr   Úultralytics.nn.tasksrt   ÚevalrC   r   Úwarningri   r   r   Úhasattr)rP   r   rk   rm   ro   r)   ÚvpeÚtpert   Ú_r<   Ú	eval_dataÚ
eval_namesr   Ú	__class__s                 €r&   ry   zYOLOEDetectValidator.get_model…   s&  ø€ õ& ‘”×!Ò! %¨Ñ1Ô1ˆØÐØ!œ.ˆDŒKØlÐlµt¸D¼OÔ<SÔ<XÐY`Ô<a×<hÒ<hÑ<jÔ<jÑ7kÔ7kÐlÑlÔlˆEàð 	.Ý”Ð?Ñ@Ô@Ð@Ø%)�”	Ô"à×(Ò(¨¬¸%Ñ@Ô@�Ø×!Ò! %¨Ñ-Ô-Ð-Ñ-å”Ð=Ñ>Ô>Ð>Ø×'Ò'¨Ñ.Ô.�Ø×!Ò! %¨Ñ-Ô-Ð-Ñ-àÐ%ØÐWÐWÐ WÑWÔW�wÝ'¨¬	Ô(8À%ÐHÑHÔHˆDŒKå˜%¥#¥t Ñ-Ô-ð FØ@Ð@Ð@Ð@Ð@Ð@à*˜?¨5¸¼ÐEÑEÔE‘��qØ�JŠJ‰LŒL�OŠO˜DœKÑ(Ô(Ð(Ý$ ZÐ%A°4´9´>ÑBÔBˆDØTÐTµt¸DÀ¼M×<PÒ<PÑ<RÔ<RÑ7SÔ7SÐTÑTÔTˆEàÐ%Ý-¨d¬i¬nÑ=Ô=�	ØbÐbÅÀYÈwÔEW×E^ÒE^ÑE`ÔE`Ñ@aÔ@aÐbÑbÔb�
Ø˜JÒ&Ð&Ý”NðJ°uð Jð JÐ_ið Jð Jð Jñô ð ð
 ð 	.Ý”Ð?Ñ@Ô@Ð@Ø%)�”	Ô"Ø!×4Ò4°TÑ:Ô:�
Ø×(Ò(¨°UÑ;Ô;�Ø×!Ò! %¨Ñ-Ô-Ð-Ð-Ý  ¤¨R¤µ+Ñ>Ô>ð .Å7È5Ì;ÐWYÌ?Ð\bÑCcÔCcð .Ý”Ð=Ñ>Ô>Ð>Ø×'Ò'¨Ñ.Ô.�Ø×!Ò! %¨Ñ-Ô-Ð-Øˆr(   )r   r   r   r   r   r   )r<   rZ   r   r   )NNNF)
r   rj   rk   rl   rm   rn   ro   rp   r   r   )	Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   rH   ri   ry   Ú__classcell__)rˆ   s   @r&   r   r      s½   ø€ € € € € ðð ð4 ÐÑÔð.&ð .&ð .&ñ Ôð.&ð`!
ð !
ð !
ð !
ðF Ð˜%Ñ Ô ð *.Ø"Ø!%ØðAð Að Að Að Að Añ !Ô ðAð Að Að Að Ar(   r   c                  ó   — e Zd ZdZdS )ÚYOLOESegValidatorzRYOLOE segmentation validator that supports both text and visual prompt embeddings.N)r‰   rŠ   r‹   rŒ   © r(   r&   r�   r�   Ê   s   € € € € € Ø\Ð\Ð\Ð\r(   r�   )#Ú
__future__r   Úpathlibr   Útypingr   r>   Útorch.nnr   rM   Úultralytics.datar   r   r	   Úultralytics.data.augmentr
   Úultralytics.data.utilsr   r   Úultralytics.models.yolo.detectr   Úultralytics.models.yolo.segmentr   Úultralytics.nn.modules.headr   r   r   Úultralytics.utilsr   r   Úultralytics.utils.torch_utilsr   r   r   r�   r�   r(   r&   ú<module>r�      sŽ  ðð #Ð "Ð "Ð "Ð "Ð "à Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð à €€€Ø $Ð $Ð $Ð $Ð $Ð $à TÐ TÐ TÐ TÐ TÐ TÐ TÐ TÐ TÐ TØ 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø HÐ HÐ HÐ HÐ HÐ HÐ HÐ HØ =Ð =Ð =Ð =Ð =Ð =Ø AÐ AÐ AÐ AÐ AÐ AØ 3Ð 3Ð 3Ð 3Ð 3Ð 3Ø +Ð +Ð +Ð +Ð +Ð +Ø *Ð *Ð *Ð *Ð *Ð *Ð *Ð *Ø MÐ MÐ MÐ MÐ MÐ MÐ MÐ Mðqð qð qð qð qÐ-ñ qô qð qðh]ð ]ð ]ð ]ð ]Ð,Ð.Cñ ]ô ]ð ]ð ]ð ]r(   