a yfZY@sbddlZddlZddlmZddlmZddlmZddlm Z ddl Z ddl Z ddl Z ddlmZddlmZddlmZmZmZmZdd lmZdd lmZd d lmZmZmZmZm Z m!Z!m"Z"m#Z#d d l$m%Z%d dl&m'Z'm(Z(m)Z)m*Z*m+Z+m,Z,m-Z-m.Z.dZ/Gddde%Z0Gddde0Z1Gddde0Z2GdddeZ3Gddde%Z4GdddZ5dS)N) defaultdict)repeat) ThreadPool)Path)Image) ConcatDataset) LOCAL_RANK NUM_THREADSTQDMcolorstr)resample_segments)TORCHVISION_0_18)ComposeFormat Instances LetterBoxRandomLoadTextclassify_augmentationsclassify_transforms v8_transforms) BaseDataset)HELP_URLLOGGERget_hashimg2label_pathsload_dataset_cache_filesave_dataset_cache_file verify_imageverify_image_labelz1.0.3csfeZdZdZdddfdd Zedfdd Zd d Zdd d ZddZ ddZ e ddZ Z S) YOLODataset Dataset class for loading object detection and/or segmentation labels in YOLO format. Args: data (dict, optional): A dataset YAML dictionary. Defaults to None. task (str): An explicit arg to point current task, Defaults to 'detect'. Returns: (torch.utils.data.Dataset): A PyTorch dataset object that can be used for training an object detection model. NdetectdatataskcsN|dk|_|dk|_|dk|_||_|jr8|jr8Jdtj|i|dS)zTInitializes the YOLODataset with optional configurations for segments and keypoints.segmentZposeobbz(Can not use both segments and keypoints.N) use_segments use_keypointsuse_obbr$super__init__selfr$r%argskwargs __class__T/var/www/html/django/DPS/env/lib/python3.9/site-packages/ultralytics/data/dataset.pyr,:s    zYOLODataset.__init__z./labels.cachec Cs4dgi}ddddgf\}}}}}|jd|j|jd}t|j} |jdd\} } |jrt| dksl| dvrttdt t  } | j t t |j|jt|jt|jtt|jd t| t| d } t| || d }|D]\ }}}}}}}}}}||7}||7}||7}||7}|rT|d|||d d dd f|d d d d f||ddd|rd|||d|d||d|d|_q|Wd n1s0Y|rtd||dkrt|jd|dtt|j|j|d<||||t|jf|d<||d<t|j||t|S)z Cache dataset labels, check images and read shapes. Args: path (Path): Path where to save the cache file. Default is Path('./labels.cache'). Returns: (dict): labels. labelsr Scanning ...Z kpt_shape)rr>z'kpt_shape' in data.yaml missing or incorrect. Should be a list with [number of keypoints, number of dims (2 for x,y or 3 for x,y,visible)], i.e. 'kpt_shape: [17, 3]'namesfunciterabledesctotalNrTxywh)im_fileshapeclsbboxessegments keypoints normalized bbox_format images,  backgrounds,  corrupt "WARNING ⚠️ No labels found in z. hashresultsmsgs)prefixparentstemlenim_filesr$getr) ValueErrorrr imaprzip label_filesrr appendr?closerinfojoinwarningrrrDATASET_CACHE_VERSION)r.pathxnmnfnencrRr?r@ZnkptndimpoolrQpbarrBlbrCrFZkeypointZnm_fnf_fZne_fnc_fmsgr3r3r4 cache_labelsCsj      $( zYOLODataset.cache_labelsc st|j|_t|jdjd}z>t|d}dtksBJdt|j|jks^JWn(t t t fy| |d}Yn0 d\}}}}}|rtdvrd |d |d ||d |d }td|j|||ddrtddfdddDd} | s:td|dtdd| D|_dd| D} ddt| D\} } } | r| | krtd| d| d| D]}g|d<q| dkrtd |dt| S)!z/Returns dictionary of labels for YOLO training.r.cacheTversionrPFrQrr6z... rKrLrMNr?r@initialrRrNcsg|]}|qSr3)pop).0kcacher3r4 z*YOLODataset.get_labels..)rPrrrRr5u"WARNING ⚠️ No images found in z#, training may not work correctly. cSsg|] }|dqS)rBr3rxrlr3r3r4r|r}css0|](}t|dt|dt|dfVqdS)rDrErFN)rVr~r3r3r4 r}z)YOLODataset.get_labels..css|]}t|VqdS)N)sumrxrdr3r3r4rr}uOWARNING ⚠️ Box and segment counts should be equal, but got len(segments) = z, len(boxes) = 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.rFrO)rrWr\rrT with_suffixrrbrFileNotFoundErrorAssertionErrorAttributeErrorrprwrr rSrr_r`rarr[)r. cache_pathexistsrfrergrhndr5lengthsZlen_clsZ len_boxesZ len_segmentsrlr3rzr4 get_labelss@  "   zYOLODataset.get_labelsc Cs|jrF|jr|js|jnd|_|jr0|js0|jnd|_t||j|}ntt|j|jfddg}|t dd|j |j |j d|j |j|jr|jndd |S)z*Builds and appends transforms to the list.F)Z new_shapeZscaleuprAT) rI normalizeZ return_maskZreturn_keypointZ return_obb batch_idx mask_ratioZ mask_overlapbgr)augmentrectmosaicmixuprimgszrrr]rr(r)r*rZ overlap_maskrr.hyp transformsr3r3r4build_transformss& zYOLODataset.build_transformscCs"d|_d|_d|_|||_dS)zLSets mosaic, copy_paste and mixup options to 0.0 and builds transformations.rN)rZ copy_pasterrr)r.rr3r3r4 close_mosaicszYOLODataset.close_mosaiccCs|d}|dg}|dd}|d}|d}|jr@dnd}t|d krhtjt||d d d }ntjd |d ftjd }t|||||d|d<|S)a Custom your label format here. Note: 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. rErFrGNrIrHdir)r)Zaxisr8Zdtype)rIrHZ instances) rwr*rVnpstackr zerosfloat32r)r.labelrErFrGrIrHZsegment_resamplesr3r3r4update_labels_infos      zYOLODataset.update_labels_infocCsi}|d}ttdd|D}t|D]@\}}||}|dkrRt|d}|dvrft|d}|||<q.t|d|d<tt|dD]}|d||7<qt|dd|d<|S)#Collates data samples into batches.rcSsg|]}t|qSr3)listvalues)rxbr3r3r4r|r}z*YOLODataset.collate_fn..img>r'rGrFrErDmasksr) keysrr[ enumeratetorchrcatrangerV)batchZ new_batchrriryvaluer3r3r4 collate_fns    zYOLODataset.collate_fn)N)__name__ __module__ __qualname____doc__r,rrprrrr staticmethodr __classcell__r3r3r1r4r .s  C) r csBeZdZdZdddfdd ZfddZd fd d ZZS) YOLOMultiModalDatasetr!Nr"r#cstj|||d|dS)zUInitializes a dataset object for object detection tasks with optional specifications.r#Nr+r,r-r1r3r4r,szYOLOMultiModalDataset.__init__cs,t|}dd|jdD|d<|S)z5Add texts information for multi-modal model training.cSsg|]\}}|dqS)/)split)rx_vr3r3r4r|r}z.r:texts)r+rr$items)r.rr5r1r3r4r s z(YOLOMultiModalDataset.update_labels_infocs6t|}|jr2|dtt|jdddd|S)zWEnhances data transformations with optional text augmentation for multi-modal training.rtrhPTZ max_samplespadding)r+rrinsertrminr$rr1r3r4rs  z&YOLOMultiModalDataset.build_transforms)N)rrrrr,rrrr3r3r1r4rs  rcsDeZdZdZddfdd ZddZdd Zd fd d ZZS)GroundingDatasetziHandles object detection tasks by loading annotations from a specified JSON file, supporting YOLO format.r")r%cs2|dksJd||_tj||id|dS)zdInitializes a GroundingDataset for object detection, loading annotations from a specified JSON file.r"z6`GroundingDataset` only support `detect` task for now!)r%r$N) json_filer+r,)r.r%rr/r0r1r3r4r,szGroundingDataset.__init__cCsgS)zOThe image files would be read in `get_labels` function, return empty list here.r3)r.img_pathr3r3r4 get_img_files#szGroundingDataset.get_img_filesc sdg}tdt|j}t|}Wdn1s80Ydd|dD}tt}|dD]}||d|qdt | d|jd D]\}}||d d d d } } }t |j |} | sq|jt| g} i} g}|D]}|drqtj|dtjd}|dd|ddd7<|ddgt| <|ddgt| <|ddks|ddkrqdfdd|dD}|| vrt| | |<||g| |}|g|}|| vr| |qt| rtj| tjdntjdtjd}|| | | f|ddddf|ddddfdd|dq|S)zZLoads annotations from a JSON file, filters, and normalizes bounding boxes for each image.zLoading annotation file...NcSsi|]}|dd|qS)idrr3rr3r3r4 -r}z/GroundingDataset.get_labels..images annotationsZimage_idzReading annotations )r?rheightwidth file_nameZiscrowdZbboxrr8rrr9rJcs$g|]}d|d|dqS)captionrrr3)rxtrr3r4r|Er}z/GroundingDataset.get_labels..Ztokens_positive)rTrA)rBrCrDrErHrIr)rr_openrjsonloadrrr]r rrrrrWstrrarrayrfloatr`rVtolistr)r.r5frrZ img_to_annsannZimg_idannshwrBrEZcat2idrboxZcat_namerDrlr3rr4r's^  ( "       * zGroundingDataset.get_labelsNcs*t|}|jr&|dtddd|S)zgConfigures augmentations for training with optional text loading; `hyp` adjusts augmentation intensity.rtrTr)r+rrrrrr1r3r4r[s z!GroundingDataset.build_transforms)N) rrrrr,rrrrr3r3r1r4rs 4rc@seZdZdZeddZdS)YOLOConcatDatasetz} Dataset as a concatenation of multiple datasets. This class is useful to assemble different existing datasets. cCs t|S)r)r r)rr3r3r4rkszYOLOConcatDataset.collate_fnN)rrrrrrr3r3r3r4rdsrcs eZdZdZfddZZS)SemanticDatasetab Semantic Segmentation Dataset. This class is responsible for handling datasets used for semantic segmentation tasks. It inherits functionalities from the BaseDataset class. Note: This class is currently a placeholder and needs to be populated with methods and attributes for supporting semantic segmentation tasks. cstdS)z$Initialize a SemanticDataset object.Nrr.r1r3r4r,~szSemanticDataset.__init__)rrrrr,rr3r3r1r4rrs rc@s8eZdZdZdddZddZedd d Zd d Zd S)ClassificationDataseta< Extends torchvision ImageFolder to support YOLO classification tasks, offering 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. This class allows for augmentations using both torchvision and Albumentations libraries, and supports caching images in RAM or on disk to reduce IO overhead during training. Additionally, it implements a robust verification process to ensure data integrity and consistency. 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 tuples, 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. Fc CsBddl}tr |jj|dd|_n|jj|d|_|jj|_|jj|_|rr|jdkrr|jdtt |j|j|_|rt |dnd|_ |j dupt |j d k|_|jrtd d |_t |j d k|_||_d d|jD|_d|jdf}|r,t|j||j|j|j|j|j|j|jd nt|j|jd|_ dS)a Initialize YOLO object with root, image size, 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. It includes attributes like `imgsz` (image size), `fraction` (fraction of data to use), `scale`, `fliplr`, `flipud`, `cache` (disk or RAM caching for faster training), `auto_augment`, `hsv_h`, `hsv_s`, `hsv_v`, and `crop_fraction`. augment (bool, optional): Whether to apply augmentations to the dataset. Default is False. prefix (str, optional): Prefix for logging and cache filenames, aiding in dataset identification and debugging. Default is an empty string. rNT)rootZ allow_empty)rg?z: rramuWARNING ⚠️ Classification `cache_ram` training has known memory leak in https://github.com/ultralytics/ultralytics/issues/9824, setting `cache_ram=False`.FZdiskcSs*g|]"}t|t|dddgqS)rz.npyN)rrrrr3r3r4r|r}z2ClassificationDataset.__init__..) sizescaleZhflipZvfliperasing auto_augmenthsv_hhsv_shsv_v)r crop_fraction)! torchvisionr ZdatasetsZ ImageFolderbasesamplesrfractionroundrVr rSr{rlower cache_ramrra cache_disk verify_imagesrrrZfliplrZflipudrrrrrrrtorch_transforms)r.rr/rrSrrr3r3r4r,sB      zClassificationDataset.__init__cCs|j|\}}}}|jr:|durxt|}|j|d<n>|jrn|sbtj|t|ddt |}n t|}t t |tj }||}||dS)zBReturns subset of data and targets corresponding to given indices.Nr9F)Z allow_pickle)rrD)rrcv2Zimreadrrrsaveas_posixrrZ fromarrayZcvtColorZ COLOR_BGR2RGBr)r.rrjfnZimsampler3r3r4 __getitem__s   z!ClassificationDataset.__getitem__)returncCs t|jS)z2Return the total number of samples in the dataset.)rVrrr3r3r4__len__szClassificationDataset.__len__cCs|jd|jd}t|jd}ttttt |}|dt ksLJ|dt dd|j DkslJ| d\}}}}td vr|d |d |d }td |||d|drtd|d|Wd S1s0Yddggif\}}} }} tt} | jtt|j t|jd} t| |t|j d} | D]V\}}}}|r\|||rl| |||7}||7}|d |d |d | _q@| Wd n1s0Y| rtd| t dd|j D| d<||t||f| d<| | d<t|j|| t |S)zVerify all images in dataset.r6r7rqrrrPcSsg|] }|dqSrr3rr3r3r4r|r}z7ClassificationDataset.verify_images..rQrsrJrKrMNrurRrNrr;r>cSsg|] }|dqSrr3rr3r3r4r|r})rSrrr contextlibsuppressrrrrrbrrrwrr rr_r`rr rZrr[rrVr]r?r^r)r.r?rcr{rfrhrrrrRrdrjrQrkrrmrnror3r3r4rsB "   (z#ClassificationDataset.verify_imagesN)Fr) rrrrr,rintrrr3r3r3r4rs  7r)6rr collectionsr itertoolsrZmultiprocessing.poolrpathlibrrnumpyrrZPILrZtorch.utils.datarZultralytics.utilsrr r r Zultralytics.utils.opsr Zultralytics.utils.torch_utilsr rrrrrrrrrrrutilsrrrrrrrrrbr rrrrrr3r3r3r4s0        ( ( M J