|h%ddlmZmZmZmZmZddlZddlmZddl m Z m Z m Z ddl mZddlmZmZdZGdd eZGd d eZy) )AnyDictListTupleUnionN) YOLODataset)ComposeFormat v8_transforms)DetectionValidator)colorstrops)RTDETRValidatorc<eZdZdZddfd Zdfd ZddZxZS) RTDETRDataseta Real-Time DEtection and TRacking (RT-DETR) dataset class extending the base YOLODataset class. This specialized dataset class is designed for use with the RT-DETR object detection model and is optimized for real-time detection and tracking tasks. Attributes: augment (bool): Whether to apply data augmentation. rect (bool): Whether to use rectangular training. use_segments (bool): Whether to use segmentation masks. use_keypoints (bool): Whether to use keypoint annotations. imgsz (int): Target image size for training. Methods: load_image: Load one image from dataset index. build_transforms: Build transformation pipeline for the dataset. Examples: Initialize an RT-DETR dataset >>> dataset = RTDETRDataset(img_path="path/to/images", imgsz=640) >>> image, hw = dataset.load_image(0) N)datac(t||d|i|y)aU Initialize the RTDETRDataset class by inheriting from the YOLODataset class. This constructor sets up a dataset specifically optimized for the RT-DETR (Real-Time DEtection and TRacking) model, building upon the base YOLODataset functionality. Args: *args (Any): Variable length argument list passed to the parent YOLODataset class. data (dict | None): Dictionary containing dataset information. If None, default values will be used. **kwargs (Any): Additional keyword arguments passed to the parent YOLODataset class. rN)super__init__)selfrargskwargs __class__s \/var/www/html/test/engine/venv/lib/python3.12/site-packages/ultralytics/models/rtdetr/val.pyrzRTDETRDataset.__init__'s $4T4V4c&t|||S)a" Load one image from dataset index 'i'. Args: i (int): Index of the image to load. rect_mode (bool, optional): Whether to use rectangular mode for batch inference. Returns: im (torch.Tensor): The loaded image. resized_hw (tuple): Height and width of the resized image with shape (2,). Examples: Load an image from the dataset >>> dataset = RTDETRDataset(img_path="path/to/images") >>> image, hw = dataset.load_image(0) )i rect_mode)r load_image)rrrrs rrzRTDETRDataset.load_image5s"w!A!;;rc |jr|jr|js |jnd|_|jr|js |jnd|_|jr|js |jnd|_t ||j |d}n tg}|jtdd|j|jd|j|j|S)z Build transformation pipeline for the dataset. Args: hyp (dict, optional): Hyperparameters for transformations. Returns: (Compose): Composition of transformation functions. gT)stretchxywh) bbox_format normalize return_maskreturn_keypoint batch_idx mask_ratio mask_overlap)augmentrectmosaicmixupcutmixr imgszr appendr use_segments use_keypointsr( overlap_mask)rhyp transformss rbuild_transformszRTDETRDataset.build_transformsHs <<'+||DII3CJ%)\\$)) CI'+||DII3CJ&tTZZdKJ!J " -- $ 2 2>> --  r)F)N)__name__ __module__ __qualname____doc__rrr6 __classcell__)rs@rrrs.$( 5<&rrc"eZdZdZd dZdeejeeje ejfdee e ejffdZ de de e efde e effd Zd e e ejfd e e efde e ejffd Zy)ra RTDETRValidator extends the DetectionValidator class to provide validation capabilities specifically tailored for the RT-DETR (Real-Time DETR) object detection model. The class allows building of an RTDETR-specific dataset for validation, applies Non-maximum suppression for post-processing, and updates evaluation metrics accordingly. Attributes: args (Namespace): Configuration arguments for validation. data (dict): Dataset configuration dictionary. Methods: build_dataset: Build an RTDETR Dataset for validation. postprocess: Apply Non-maximum suppression to prediction outputs. Examples: Initialize and run RT-DETR validation >>> from ultralytics.models.rtdetr import RTDETRValidator >>> args = dict(model="rtdetr-l.pt", data="coco8.yaml") >>> validator = RTDETRValidator(args=args) >>> validator() Notes: For further details on the attributes and methods, refer to the parent DetectionValidator class. Nc t||jj|d|jd|jjxsdt |d|j  S)a Build an RTDETR Dataset. Args: img_path (str): Path to the folder containing images. mode (str, optional): `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: (RTDETRDataset): Dataset configured for RT-DETR validation. FNz: ) img_pathr/ batch_sizer*r4r+cacheprefixr)rrr/r@r r)rr>modebatchs r build_datasetzRTDETRValidator.build_datasetsT))// ))//)TtfBK(  rpredsreturnct|ttfs|dg}|dj\}}}|dj d|dz fd\}}||j j z}tjd|jg|z}t|D]\}} tj| } ||jd\} } tj| | d| dgd} | | jd } | | |j j kD||<|D cgc]} | ddddf| dddf| ddd fd !c} Scc} w) a Apply Non-maximum suppression to prediction outputs. Args: preds (torch.Tensor | List | Tuple): Raw predictions from the model. If tensor, should have shape (batch_size, num_predictions, num_classes + 4) where last dimension contains bbox coords and class scores. Returns: (List[Dict[str, torch.Tensor]]): List of dictionaries for each image, each containing: - 'bboxes': Tensor of shape (N, 4) with bounding box coordinates - 'conf': Tensor of shape (N,) with confidence scores - 'cls': Tensor of shape (N,) with class indices Nr)dim)r)device).NT) descendingbboxesconfcls) isinstancelisttupleshapesplitrr/torchzerosrL enumerater xywh2xyxymaxcatargsortrQ)rrEbs_ndrPscoresoutputsrbboxscorerRpredxs r postprocesszRTDETRValidator.postprocesssE %$/DME!HNN ArqBF <$))//!;;vfmm<=B ( 6GAt==&Dr*JE399dE)$4c)nE2ND  67Dediinn45GAJ  6PWW!1QU8Qq!tWQq!tWEWWWs!$EsirCc4|d|k(}|d|jd}|d|}|d|}|djdd}|d |}t|r=tj|}|d d dgfxx|d zcc<|d d d gfxx|d zcc<|||||dS)a Prepare a batch for validation by applying necessary transformations. Args: si (int): Batch index. batch (Dict[str, Any]): Batch data containing images and annotations. Returns: (Dict[str, Any]): Prepared batch with transformed annotations containing cls, bboxes, ori_shape, imgsz, and ratio_pad. r'rRrIrP ori_shapeimgN ratio_pad.r)rRrPrkr/rn)squeezerVlenrr[) rrirCidxrRrdrkr/rns r_prepare_batchzRTDETRValidator._prepare_batchsK B&El3''+Xs#+&r* e ""12&+&r* s8==&D q!f 1 -  q!f 1 - dUajkkrrfpbatchc2|d}|jjr|dz}|dj}|dddgfxx|dd|jjz zcc<|dddgfxx|dd|jjz zcc<||d |d S) a Prepare predictions by scaling bounding boxes to original image dimensions. Args: pred (Dict[str, torch.Tensor]): Raw predictions containing 'cls', 'bboxes', and 'conf'. pbatch (Dict[str, torch.Tensor]): Prepared batch information containing 'ori_shape' and other metadata. Returns: (Dict[str, torch.Tensor]): Predictions scaled to original image dimensions. rRrrP.rmrkrorprQrO)r single_clscloner/)rrfrurRrPs r _prepare_predzRTDETRValidator._prepare_preds5k 99   1HCh%%'sQF{vk215 GGsQF{vk215 GG $v,sCCr)valN)r7r8r9r:rDrrXTensorrrrstrrhintrrtryrrrrhs4 2X5<<ell);U5<<=PPQX d3 $% &XBllT#s(^lS#Xl0D$sELL'8"9D4S>DVZ[^`e`l`l[lVmDrr)typingrrrrrrXultralytics.datarultralytics.data.augmentr r r ultralytics.models.yolo.detectr ultralytics.utilsr r__all__rrr~rrrsD10 (CC=+ VKVr~D(~Dr