/j%ddlmZddlmZddlmZddlZddlmZmZddl m Z ddl m Z ddl mZdd lmZmZmZdd lmZdd lmZmZGd d e ZdS)) annotations)copy)AnyN)ClassificationDatasetbuild_dataloader) BaseTrainer)yolo)ClassificationModel) DEFAULT_CFGLOGGERRANK) plot_images) is_paralleltorch_distributed_zero_firstceZdZdZeddfd&fd ZdZd'd(d Zfd Zd)d*dZ d+d,dZ d-dZ d.dZ dZ d/d0d#Zd1d%ZxZS)2ClassificationTraineraA trainer class extending BaseTrainer for training image classification models. This trainer handles the training process for image classification tasks, supporting both YOLO classification models and torchvision models with comprehensive dataset handling and validation. Attributes: model (ClassificationModel): The classification model to be trained. data (dict[str, Any]): Dictionary containing dataset information including class names and number of classes. loss_names (list[str]): Names of the loss functions used during training. validator (ClassificationValidator): Validator instance for model evaluation. Methods: set_model_attributes: Set the model's class names from the loaded dataset. get_model: Return a modified PyTorch model configured for training. setup_model: Load, create or download model for classification. build_dataset: Create a ClassificationDataset instance. get_dataloader: Return PyTorch DataLoader with transforms for image preprocessing. preprocess_batch: Preprocess a batch of images and classes. progress_string: Return a formatted string showing training progress. get_validator: Return an instance of ClassificationValidator. label_loss_items: Return a loss dict with labeled training loss items. final_eval: Evaluate trained model and save validation results. plot_training_samples: Plot training samples with their annotations. Examples: Initialize and train a classification model >>> from ultralytics.models.yolo.classify import ClassificationTrainer >>> args = dict(model="yolo26n-cls.pt", data="imagenet10", epochs=3) >>> trainer = ClassificationTrainer(overrides=args) >>> trainer.train() N overridesdict[str, Any] | None _callbacks dict | Nonec|i}d|d<|dd|d<t|||dS)aInitialize a ClassificationTrainer object. Args: cfg (dict[str, Any], optional): Default configuration dictionary containing training parameters. overrides (dict[str, Any], optional): Dictionary of parameter overrides for the default configuration. _callbacks (dict, optional): Dictionary of callback functions to be executed during training. Nclassifytaskimgsz)getsuper__init__)selfcfgrr __class__s k/home/longshao/multi-rider-rag/.venv/lib/python3.11/site-packages/ultralytics/models/yolo/classify/train.pyrzClassificationTrainer.__init__4sX  I& & == ! ! )!$Ig  i44444c4|jd|j_dS)z9Set the YOLO model's class names from the loaded dataset.namesN)datamodelr%rs r"set_model_attributesz*ClassificationTrainer.set_model_attributesCs9W- r#Tverboseboolct||jd|jd|o tdk}|r|||D]n}|jjs$t|dr|t|tj j r|jj r|jj |_o|D] }d|_ |S)aReturn a modified PyTorch model configured for training YOLO classification. Args: cfg (Any, optional): Model configuration. weights (Any, optional): Pre-trained model weights. verbose (bool, optional): Whether to display model information. Returns: (ClassificationModel): Configured PyTorch model for classification. ncchannels)r-chr*reset_parametersT)r r&r loadmodulesargs pretrainedhasattrr1 isinstancetorchnnDropoutdropoutp parameters requires_grad)rr weightsr*r'mr<s r" get_modelzClassificationTrainer.get_modelGs$CDIdO *@U_f_ukosukuvvv  JJw    ( (A9' %GA7I,J,J %""$$$!UX-.. (493D (i'!!## # #A"AOO r#cHddl}t|j|jjvr7|jj|j|jjrdnd|_d}n t}tj |j|j d|S)zLoad, create or download model for classification tasks. Returns: (Any): Model checkpoint if applicable, otherwise None. rN IMAGENET1K_V1)r?r-) torchvisionstrr'models__dict__r4r5r setup_modelr reshape_outputsr&)rrDckptr!s r"rHz!ClassificationTrainer.setup_model_s  tz??k09 9 9@+4TZ@+/9+?ITDJDD77&&((D+DJ $HHH r#trainimg_pathrEmodec:t||j|dk|S)aCreate a ClassificationDataset instance given an image path and mode. Args: img_path (str): Path to the dataset images. mode (str, optional): Dataset mode ('train', 'val', or 'test'). batch (Any, optional): Batch information (unused in this implementation). Returns: (ClassificationDataset): Dataset for the specified mode. rK)rootr4augmentprefix)rr4)rrLrMbatchs r" build_datasetz#ClassificationTrainer.build_datasetqs#%(DT[Odhiiiir#r dataset_path batch_sizeintrankc  t|5|||}dddn #1swxYwY|jdd t |jj} r| kry|jj d}t |j} fd|jD|_|t |jz } tj |d|d d| d| t|||j j ||j j } |d krFt|jr| jj|jj_n| jj|j_| S) aReturn PyTorch DataLoader with transforms to preprocess images. Args: dataset_path (str): Path to the dataset. batch_size (int, optional): Number of images per batch. rank (int, optional): Process rank for distributed training. mode (str, optional): 'train', 'val', or 'test' mode. Returns: (torch.utils.data.DataLoader): DataLoader for the specified dataset and mode. Nr-rc,g|]}|dk|S)).0sr-s r" z8ClassificationTrainer.get_dataloader..s"GGGQQqTBYYqYYYr#z split has z classes but model expects z . Skipping z samples from extra classes: )rX drop_lastrK)rrSr&rlenbaseclassessamplesr warningrr4workerscompilerr'datasettorch_transformsmodule transforms) rrUrVrXrMrh dataset_nc extra_classesoriginal_countskippedloaderr-s @r"get_dataloaderz$ClassificationTrainer.get_dataloader~s*$ / / = =((t<.s#999A6A999r#N)rroundfloatdictzip)rrrQkeyss ` r"label_loss_itemsz&ClassificationTrainer.label_loss_itemss]:999999  KE*--q112 Cj))***r#nictj|djd|d<t||jd|dz |jdS)zPlot training samples with their annotations. Args: batch (dict[str, torch.Tensor]): Batch containing images and class labels. ni (int): Batch index used for naming the output file. rur batch_idx train_batchz.jpg)labelsfnameon_plotN)r8arangeshaperrr)rrRrs r"plot_training_samplesz+ClassificationTrainer.plot_training_samplessc#\%,*rs#""""" DDDDDDDD222222######4444447777777777222222SSSSSSSSC C C C C KC C C C C r#