Fi-.ddlmZddlZddlZddlmZddlmZddlZddl Z ddl m Z ddl m Z mZddlmZddlmZddlmZdd lmZmZmZdd lmZdd lmZmZdd lmZm Z Gd deZ!y)) annotationsN)copy)Any)build_dataloaderbuild_yolo_dataset) BaseTrainer)yolo)DetectionModel) DEFAULT_CFGLOGGERRANK)override_configs) plot_images plot_labels)torch_distributed_zero_first unwrap_modelceZdZdZeddfdfd ZdddZdddZddZdZ dZ ddd Z d Z ddd Z d Zdd ZdZfdZxZS)DetectionTraineraA class extending the BaseTrainer class for training based on a detection model. This trainer specializes in object detection tasks, handling the specific requirements for training YOLO models for object detection including dataset building, data loading, preprocessing, and model configuration. Attributes: model (DetectionModel): The YOLO detection model being trained. data (dict): Dictionary containing dataset information including class names and number of classes. loss_names (tuple): Names of the loss components used in training (box_loss, cls_loss, dfl_loss). Methods: build_dataset: Build YOLO dataset for training or validation. get_dataloader: Construct and return dataloader for the specified mode. preprocess_batch: Preprocess a batch of images by scaling and converting to float. set_model_attributes: Set model attributes based on dataset information. get_model: Return a YOLO detection model. get_validator: Return a validator for model evaluation. label_loss_items: Return a loss dictionary with labeled training loss items. progress_string: Return a formatted string of training progress. plot_training_samples: Plot training samples with their annotations. plot_training_labels: Create a labeled training plot of the YOLO model. auto_batch: Calculate optimal batch size based on model memory requirements. Examples: >>> from ultralytics.models.yolo.detect import DetectionTrainer >>> args = dict(model="yolo26n.pt", data="coco8.yaml", epochs=3) >>> trainer = DetectionTrainer(overrides=args) >>> trainer.train() Nc(t||||y)aInitialize a DetectionTrainer object for training YOLO object detection models. Args: cfg (dict, optional): Default configuration dictionary containing training parameters. overrides (dict, optional): Dictionary of parameter overrides for the default configuration. _callbacks (dict, optional): Dictionary of callback functions to be executed during training. N)super__init__)selfcfg overrides _callbacks __class__s f/var/www/pod-logistic/pod-ai/venv/lib/python3.12/site-packages/ultralytics/models/yolo/detect/train.pyrzDetectionTrainer.__init__7s i4c ttt|jjjd}t |j |||j||dk(|S)aBuild 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' mode. Returns: (Dataset): YOLO dataset object configured for the specified mode. val)moderectstride)maxintrmodelr$rargsdata)rimg_pathr"batchgss r build_datasetzDetectionTrainer.build_datasetAsT\$**-4488:;R @!$))XudiidY]afYfoqrrrc |dvs Jd|dt|5|j|||}ddd|dk(}tddrH|rFtj|j |j dk(st jd d}t|||dk(r|jjn|jjd z|||jjxr|dk( S#1swYxYw) aConstruct and return dataloader for the specified mode. Args: dataset_path (str): Path to the dataset. batch_size (int): Number of images per batch. rank (int): Process rank for distributed training. mode (str): 'train' for training dataloader, 'val' for validation dataloader. Returns: (DataLoader): PyTorch dataloader object. >r!trainz#Mode must be 'train' or 'val', not .Nr/r#FrzJ'rect=True' is incompatible with DataLoader shuffle, setting shuffle=False)r+workersshufflerank drop_last) rr-getattrnpall batch_shapesr warningrr(r2compile)r dataset_path batch_sizer4r"datasetr3s rget_dataloaderzDetectionTrainer.get_dataloaderOs''V+NtfTU)VV' )$ / I((tZHG I'/ 7FE *wrvvgFZFZ^e^r^rst^uFu?v NNg hG )-DII%%dii>O>ORS>Sii'';DGO    I Is C//C8c|jD]W\}}t|tjs!|j |j |j j dk(||<Y|djdz |d<|jjdkDrD|d}tjt|jjd|jjz zt|jjd|jjzz|jz|jz|jz}|t|j ddz }|d k7ro|j ddDcgc]4}t#j$||z|jz |jz6}}t&j(j+||d d }||d<|Scc}w) zPreprocess a batch of images by scaling and converting to float. Args: batch (dict): Dictionary containing batch data with 'img' tensor. Returns: (dict): Preprocessed batch with normalized images. cuda) non_blockingimg?r1NbilinearF)sizer" align_corners)items isinstancetorchTensortodevicetypefloatr( multi_scalerandom randranger&imgszr$r%shapemathceilnn functional interpolate) rr+kvimgsszsfxnss rpreprocess_batchz!DetectionTrainer.preprocess_batchksKKM VDAq!U\\*44 $++:J:Jf:T4Ua VU|))+c1e 99 3 &FGDIIa"ft{{23dkkA}}00BZ_d0eE%L s?9G$c@|jd|j_|jd|j_|j|j_t |jdr1|jj |jjyy)z2Set model attributes based on dataset information.ncnamesend2end)max_detN)r)r'rfrgr(r6 set_head_attrrirs rset_model_attributesz%DetectionTrainer.set_model_attributessm  $ 99W- ))  4::y ) JJ $ $TYY->-> $ ? *rcd|jjcxkr dksJdJd|jjdk(rytj|jj j Dcgc]}|djc}d}tj|jt|jdjtj}tj|dk(d|}d|z |jjz}||jz }tj |j#|j$|j&_t+j,d |j&j(j/j1j3d ycc}w) a8Compute and set class weights for handling class imbalance. Class weights are computed based on inverse class frequency in the training dataset, raised to the power of cls_pw (0 < cls_pw <= 1 dampens, cls_pw > 1 amplifies). Final weights are normalized so their mean equals 1.0. rrFz"cls_pw must be in the range [0, 1]rENclsrf) minlengthzClass weights: )r(cls_pwr7 concatenate train_loaderr>labelsflattenbincountastyper&r)float32wheremeanrM from_numpyrOrPr' class_weightsr infocpunumpyround)rlbclasses class_countsweightss rset_class_weightsz"DetectionTrainer.set_class_weightss^DII$$++Q-QQ+Q-QQ+ 99  s " ..@Q@Q@Y@Y@`@`!a""U)"3"3"5!acde{{7>>##6$))D/RYYZ\ZdZde xx  13 E %$))*:*::GLLN*#(#3#3G#<#?#? #L   odjj&>&>&B&B&D&J&J&L&R&RST&U%VWX"bs:Gct||jd|jd|xr tdk(}|r|j||S)a=Return a YOLO detection model. Args: cfg (str, optional): Path to model configuration file. weights (str, optional): Path to model weights. verbose (bool): Whether to display model information. Returns: (DetectionModel): YOLO detection model. rfchannels)rfchverbose)r r)r load)rrrrr's r get_modelzDetectionTrainer.get_modelsFstyy499Z;PZaZpfjnpfpq  JJw  rcd|_tjj|j|j t |j|jS)z6Return a DetectionValidator for YOLO model validation.)box_losscls_lossdfl_loss)save_dirr(r) loss_namesr detectDetectionValidator test_loaderrrr( callbacksrks r get_validatorzDetectionTrainer.get_validatorsG<{{--   t}}4 ?W[WeWe.  rc|jDcgc] }|d| }}|7|Dcgc]}tt|d}}tt ||S|Scc}wcc}w)a_Return a loss dict with labeled training loss items tensor. Args: loss_items (list[float], optional): List of loss values. prefix (str): Prefix for keys in the returned dictionary. Returns: (dict | list): Dictionary of labeled loss items if loss_items is provided, otherwise list of keys. /)rrrRdictzip)r loss_itemsprefixrbkeyss rlabel_loss_itemsz!DetectionTrainer.label_loss_itemssh*.9A6(!A399  !6@A%a!,AJAD*-. .K :As AAcjdddt|jzzzddg|jddzS)z`Return a formatted string of training progress with epoch, GPU memory, loss, instances and size. z%11sEpochGPU_mem InstancesSize)lenrrks rprogress_stringz DetectionTrainer.progress_stringsTvS%9!9::  ? __?   ?  ?   rc^t||d|jd|dz |jy)zPlot training samples with their annotations. Args: batch (dict[str, Any]): Dictionary containing batch data. ni (int): Batch index used for naming the output file. im_file train_batchz.jpg)rtpathsfnameon_plotN)rrr)rr+nis rplot_training_samplesz&DetectionTrainer.plot_training_sampless3  "--Kt4"88LL  rctj|jjjDcgc]}|d c}d}tj|jjjDcgc]}|d c}d}t ||j |jd|j|jycc}wcc}w)z1Create a labeled training plot of the YOLO model.bboxesrrnrg)rgrrN) r7rrrsr>rtrsqueezer)rr)rrboxesrns rplot_training_labelsz%DetectionTrainer.plot_training_labelsst7H7H7P7P7W7WX8 XZ[\nn$2C2C2K2K2R2RSBbiSUVWE3;;= '0BT]]dhdpdpq YSs C 7 Cc$t|jddi5|_|j|jddd}dddt dj Dd z}t |}~t|!|| S#1swYExYw) zGet optimal batch size by calculating memory occupation of model. Returns: (int): Optimal batch size. cacheF)rr/)r"r+Nc38K|]}t|dyw)rnN)r).0labels r z.DetectionTrainer.auto_batch..sN#eEl+Nsr) dataset_size) rr(r-r)r%rtrr auto_batch)r train_dataset max_num_objnrs rrzDetectionTrainer.auto_batchs diiGU3C D [  ..tyy/AWY.ZM [N9M9MNNQRR   w!+A!>>  [ [s 'BB)rzdict[str, Any] | Nonerz dict | None)r/N)r*strr"rr+z int | None)rrr/)r<rr=r&r4r&r"r)r+rreturnr)NNT)r str | Nonerrrbool)Nr/)rzlist[float] | Nonerr)r+zdict[str, Any]rr&rNone)__name__ __module__ __qualname____doc__r rr-r?rdrlrrrrrrrr __classcell__)rs@rrrs]<'4ko5 s 8@ @Y&  "   r ? ?rr)" __future__rrXrTrtypingrrr7rMtorch.nnrZultralytics.datarrultralytics.engine.trainerrultralytics.modelsr ultralytics.nn.tasksr ultralytics.utilsr r r ultralytics.utils.patchesrultralytics.utils.plottingrrultralytics.utils.torch_utilsrrrrrrsH#  A2#/776?Tc?{c?r