o W h @slddlZddlmZddlmZddlmZddlmZm Z m Z ddl m Z ddZ Gd d d ejjZdS) N)build_yolo_dataset)yolo) WorldModel) DEFAULT_CFGRANKchecks) de_parallelcCstdvrddt|jjjdD}t|jjj|ddt |j j }|j jd|d\|_}|j D]}|dq7d S) zUCallback to set up model classes and text encoder at the end of the pretrain routine.>rcSsg|] }|ddqS)/r)split).0namerW/var/www/vscode/kcb/lib/python3.10/site-packages/ultralytics/models/yolo/world/train.py sz+on_pretrain_routine_end..namesF)cache_clip_modelzViT-B/32)deviceN)rlist test_loaderdatasetdatavaluesrema set_classesnextmodel parametersrclipload text_modelrequires_grad_)trainerrr_prrron_pretrain_routine_end s  r%csHeZdZdZeddffdd Zd ddZdd d Zfd d ZZ S) WorldTrainera@ A class to fine-tune a world model on a close-set dataset. This trainer extends the DetectionTrainer to support training YOLO World models, which combine visual and textual features for improved object detection and understanding. Attributes: clip (module): The CLIP module for text-image understanding. text_model (module): The text encoder model from CLIP. model (WorldModel): The YOLO World model being trained. data (dict): Dataset configuration containing class information. args (dict): Training arguments and configuration. Examples: >>> from ultralytics.models.yolo.world import WorldModel >>> args = dict(model="yolov8s-world.pt", data="coco8.yaml", epochs=3) >>> trainer = WorldTrainer(overrides=args) >>> trainer.train() NcsX|duri}t|||zddl}Wnty&tdddl}Ynw||_dS)a Initialize a WorldTrainer object with given arguments. Args: cfg (dict): Configuration for the trainer. overrides (dict, optional): Configuration overrides. _callbacks (list, optional): List of callback functions. Nrz+git+https://github.com/ultralytics/CLIP.git)super__init__r ImportErrorrcheck_requirements)selfcfg overrides _callbacksr __class__rrr(-s      zWorldTrainer.__init__TcCsZtt|tr |dn||jdt|jdd|otdkd}|r%|||dt|S)a^ Return WorldModel initialized with specified config and weights. Args: cfg (Dict | str, optional): Model configuration. weights (str, optional): Path to pretrained weights. verbose (bool): Whether to display model info. Returns: (WorldModel): Initialized WorldModel. yaml_filechannelsncPr )chr3verboser%) r isinstancedictrminrr add_callbackr%)r+r,weightsr6rrrr get_modelBs   zWorldTrainer.get_modeltrainc CsHtt|jr t|jjndd}t|j|||j||dk||dkdS)a Build YOLO Dataset for training or validation. Args: img_path (str): Path to the folder containing images. mode (str): `train` mode or `val` mode, users are able to customize different augmentations for each mode. batch (int, optional): Size of batches, this is for `rect`. Returns: (Dataset): YOLO dataset configured for training or validation. r valr=)moderectstride multi_modal)maxintrrrBrargsr)r+img_pathr@batchgsrrr build_dataset\s$ zWorldTrainer.build_datasetcst|}ttj|d}|j||dj}|j |j|dj d}||j dddd}| t|dd|jd|d<|S) z=Preprocess a batch of images and text for YOLOWorld training.textsimg)dtyper T)r$dimkeepdim txt_feats)r'preprocess_batchr itertoolschainrtokenizetorr encode_textrMnormreshapelenshape)r+rHrK text_tokenrQr/rrrRms  zWorldTrainer.preprocess_batch)NNT)r=N) __name__ __module__ __qualname____doc__rr(r<rJrR __classcell__rrr/rr&s   r&)rSultralytics.datarultralytics.modelsrultralytics.nn.tasksrultralytics.utilsrrrultralytics.utils.torch_utilsrr%detectDetectionTrainerr&rrrrs