Fi [dZddlmZddlZddlZddlZddlmZddlm Z m Z ddl Z ddl Z ddlZddlmZmZddlmZddlmZdd lmZdd lmZmZmZmZmZmZmZdd l m!Z!m"Z"dd l#m$Z$dd l%m&Z&m'Z'm(Z(dZ)GddZ*y)a Run prediction on images, videos, directories, globs, YouTube, webcam, streams, etc. Usage - sources: $ yolo mode=predict model=yolo26n.pt source=0 # webcam img.jpg # image vid.mp4 # video screen # screenshot path/ # directory list.txt # list of images list.streams # list of streams 'path/*.jpg' # glob 'https://youtu.be/LNwODJXcvt4' # YouTube 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP, TCP stream Usage - formats: $ yolo mode=predict model=yolo26n.pt # PyTorch yolo26n.torchscript # TorchScript yolo26n.onnx # ONNX Runtime or OpenCV DNN with dnn=True yolo26n_openvino_model # OpenVINO yolo26n.engine # TensorRT yolo26n.mlpackage # CoreML (macOS-only) yolo26n_saved_model # TensorFlow SavedModel yolo26n.pb # TensorFlow GraphDef yolo26n.tflite # TensorFlow Lite yolo26n_edgetpu.tflite # TensorFlow Edge TPU yolo26n_paddle_model # PaddlePaddle yolo26n.mnn # MNN yolo26n_ncnn_model # NCNN yolo26n_imx_model # Sony IMX yolo26n_rknn_model # Rockchip RKNN yolo26n_executorch_model # PyTorch Executorch yolo26n_axelera_model # Axelera AI ) annotationsN)Path)AnyCallable)get_cfg get_save_dir)load_inference_source) LetterBox) AutoBackend) DEFAULT_CFGLOGGERMACOSWINDOWS callbackscolorstrops) check_imgsz check_imshow)increment_path)attempt_compile select_devicesmart_inference_modea Inference results will accumulate in RAM unless `stream=True` is passed, which can cause out-of-memory errors for large sources or long-running streams and videos. See https://docs.ultralytics.com/modes/predict/ for help. Example: results = model(source=..., stream=True) # generator of Results objects for r in results: boxes = r.boxes # Boxes object for bbox outputs masks = r.masks # Masks object for segment masks outputs probs = r.probs # Class probabilities for classification outputs ceZdZdZeddf ddZddZddZddZdZ dddZ dd Z ddd Z e dd Zddd Zdd ZdddZd d!dZd"dZd#dZy)$ BasePredictoraA base class for creating predictors. This class provides the foundation for prediction functionality, handling model setup, inference, and result processing across various input sources. Attributes: args (SimpleNamespace): Configuration for the predictor. save_dir (Path): Directory to save results. done_warmup (bool): Whether the predictor has finished setup. model (torch.nn.Module): Model used for prediction. data (str): Data configuration. device (torch.device): Device used for prediction. dataset (Dataset): Dataset used for prediction. vid_writer (dict[Path, cv2.VideoWriter]): Dictionary of {save_path: video_writer} for saving video output. plotted_img (np.ndarray): Last plotted image. source_type (SimpleNamespace): Type of input source. seen (int): Number of images processed. windows (list[str]): List of window names for visualization. batch (tuple): Current batch data. results (list[Any]): Current batch results. transforms (Callable): Image transforms for classification. callbacks (dict[str, list[Callable]]): Callback functions for different events. txt_path (Path): Path to save text results. _lock (threading.Lock): Lock for thread-safe inference. Methods: preprocess: Prepare input image before inference. inference: Run inference on a given image. postprocess: Process raw predictions into structured results. predict_cli: Run prediction for command line interface. setup_source: Set up input source and inference mode. stream_inference: Stream inference on input source. setup_model: Initialize and configure the model. write_results: Write inference results to files. save_predicted_images: Save prediction visualizations. show: Display results in a window. run_callbacks: Execute registered callbacks for an event. add_callback: Register a new callback function. Nct|||_t|j|_|jjd|j_d|_|jj rtd|j_d|_|jj|_ d|_ d|_ d|_ i|_ d|_d|_d|_g|_d|_d|_d|_|xst+j,|_d|_t1j2|_t+j6|y)a:Initialize the BasePredictor class. Args: cfg (str | Path | dict | SimpleNamespace): Path to a configuration file or a configuration dictionary. overrides (dict, optional): Configuration overrides. _callbacks (dict, optional): Dictionary of callback functions. Ng?FT)warnr)rargsrsave_dirconf done_warmupshowrmodeldataimgszdevicedataset vid_writer plotted_img source_typeseenwindowsbatchresults transformsrget_default_callbackstxt_path threadingLock_lockadd_integration_callbacks)selfcfg overrides _callbackss ^/var/www/pod-logistic/pod-ai/venv/lib/python3.12/site-packages/ultralytics/engine/predictor.py__init__zBasePredictor.__init__psC+ $TYY/ 99>> !!DIIN  99>>)t4DIIN IINN        #Hy'F'F'H ^^% ++D1ct|tj }|r{tj|j |}|j ddk(r |ddddf}|jd}tj|}tj|}|j|j}|jjr|jn|j}|r|dz}|S)aPrepare input image before inference. Args: im (torch.Tensor | list[np.ndarray]): Images of shape (N, 3, H, W) for tensor, [(H, W, 3) x N] for list. Returns: (torch.Tensor): Preprocessed image tensor of shape (N, 3, H, W). .N)rr>) isinstancetorchTensornpstack pre_transformshape transposeascontiguousarray from_numpytor%r"fp16halffloat)r5im not_tensors r9 preprocesszBasePredictor.preprocesss$B 55 $,,R01Bxx|q TrT ]l+B%%b)B!!"%B UU4;; **//RWWYrxxz  #IB r;c\|jjrS|jjs=t |j t |jddjz dnd}|j|g||jj||jjd|S)zGRun inference on a given image using the specified model and arguments.rT)mkdirF)augment visualizeembed) rrVr)tensorrrrr,stemr"rUrW)r5rPrkwargsrVs r9 inferencezBasePredictor.inferencesyy""D,<,<,C,C 4==4 1 a0@+A+F+FFd S  tzz"ufjudii&7&79TXT]T]TcTcuntuur;ct|Dchc]}|jc}dk(}t|j|xre|jj xrM|j jdk(xs2t|j ddxr|j jdk7|j j}|Dcgc] }|| c}Scc}wcc}w)zPre-transform input image before inference. Args: im (list[np.ndarray]): List of images with shape [(H, W, 3) x N]. Returns: (list[np.ndarray]): List of transformed images. r?ptdynamicFimx)autostride)image) lenrHr r$rrectr"formatgetattrra)r5rPx same_shapes letterboxs r9rGzBasePredictor.pre_transformsB/q177/0A5  JJv v""d*twtzz9e/T/sY]YcYcYjYjnsYs::$$  -//q "//00s C>Cc|S)z6Post-process predictions for an image and return them.)r5predsimg orig_imgss r9 postprocesszBasePredictor.postprocesss r;c||_|r|j||g|i|St|j||g|i|S)aPerform inference on an image or stream. Args: source (str | Path | list[str] | list[Path] | list[np.ndarray] | np.ndarray | torch.Tensor, optional): Source for inference. model (str | Path | torch.nn.Module, optional): Model for inference. stream (bool): Whether to stream the inference results. If True, returns a generator. *args (Any): Additional arguments for the inference method. **kwargs (Any): Additional keyword arguments for the inference method. Returns: (list[ultralytics.engine.results.Results] | generator): Results objects or generator of Results objects. )streamstream_inferencelist)r5sourcer"rqrrZs r9__call__zBasePredictor.__call__sR (4((HHH H---feMdMfMN Nr;c6|j||}|D]}y)a<Method used for Command Line Interface (CLI) prediction. This function is designed to run predictions using the CLI. It sets up the source and model, then processes the inputs in a streaming manner. This method ensures that no outputs accumulate in memory by consuming the generator without storing results. Args: source (str | Path | list[str] | list[Path] | list[np.ndarray] | np.ndarray | torch.Tensor, optional): Source for inference. model (str | Path | torch.nn.Module, optional): Model for inference. Notes: Do not modify this function or remove the generator. The generator ensures that no outputs are accumulated in memory, which is critical for preventing memory issues during long-running predictions. N)rr)r5rtr"gen_s r9 predict_clizBasePredictor.predict_clis( ##FE2 A  r;c t|jj|xs|jjd|_t ||jj |jj|jjt|jdd|_ |jj|_ |jjsO|jjs9t|jdkDs!tt|jddgr*d d l}t|d d st#j$t&i|_y ) aSet up source and inference mode. Args: source (str | Path | list[str] | list[Path] | list[np.ndarray] | np.ndarray | torch.Tensor): Source for inference. stride (int, optional): Model stride for image size checking. r@)ramin_dimchannelsr>)rtr, vid_stridebufferr|i video_flagFrNrqT)rrr$r"rar r,r} stream_bufferrfr&r)rq screenshotrcany torchvisionr warningSTREAM_WARNINGr')r5rtrars r9 setup_sourcezBasePredictor.setup_sources!9T4::CTCT^_` ,))//yy++99**TZZQ7    <<33    # #**4<< 4'74<<w?@ 440~.r;c /N Kjjrtjdjsj |j 5j||njjjjsjjrBjjrjdz n jjddjstjjjjdvrdnj j"jj$gj&d_ d gdc___t/j0j2 t/j0j2 t/j0j2 f}j5d j D]S}|_j5d j,\}}} |d 5j7|} ddd|d5j8 g|i|} jj:r1t=| t>j@r| gn| Ed{ ddd ddd|d 5jC  |_"dddj5dtG|} tI| D]} xj(dz c_|d jJdz| z |djJdz| z |d jJdz| z djD| _&jjsCjjs-jjsjjNs| | xxjQ| tS||  | z cc< jjr$tjdjW| j5djDEd{VdddjXj[D]-}t=|t\j^s|ja/jjNrt]jbjjrj(r~tefdD}tjdtgjj,j(tijddg jjd d|zjjs,jjsjjlrtGtojjqd}jjrd|dd|dkDzdjdz nd} tjdtsdj| j5dy#1swYxYw7#1swYxYw#1swYxYw#tT$rY8wxYw7D#1swYAxYww)a+Stream inference on input source and save results to file. Args: source (str | Path | list[str] | list[Path] | list[np.ndarray] | np.ndarray | torch.Tensor, optional): Source for inference. model (str | Path | torch.nn.Module, optional): Model for inference. *args (Any): Additional arguments for the inference method. **kwargs (Any): Additional keyword arguments for the inference method. Yields: (ultralytics.engine.results.Results): Results objects. NlabelsTparentsexist_ok>r]tritonr?)r$r)r%on_predict_starton_predict_batch_startr@on_predict_postprocess_end@@)rRr[ro on_predict_batch_endc3VK|] }|jjz dz"yw)rN)tr*).0rgr5s r9 z1BasePredictor.stream_inference..ys"?accDIIo+?s&)zRSpeed: %.1fms preprocess, %.1fms inference, %.1fms postprocess per image at shape r|r>z labels/*.txtz labelsz saved to zResults saved to boldon_predict_end):rverboser infor" setup_modelr3rrtsavesave_txtrrTr warmuprer&bsr|r$r*r+r,rProfiler% run_callbacksrRr[rWrBrCrDror-rcrangedtspeedr! write_resultsr StopIterationjoinr'valuescv2 VideoWriterreleasedestroyAllWindowstupleminrfrH save_croprsglobr)r5rtr"rrZ profilersr,pathsim0srrPrlnivrnls` r9rrzBasePredictor.stream_inferencesf 99   KKOzz   U # ZZC (   (:f @P@P Qyy~~!3!3-1YY-?-?)T]]YYbfquYv## !!!ZZ..2BB  ++"$( 23R /DIt|TZ 4;;/ 4;;/ 4;;/I   1 2) (" ""#;)B+4Q>  ! ! # 99  ?Y??A KKd 3WTZZUV5WgZ\ZbZbcdceZfghjlmn  99>>TYY//4993F3FT$--,,^<=>BW[W`W`WiWi"RDsb1f~.j9Q8RSoqA KK+HVT]],K+LQCP Q +,q//Y!!EE %(GC (C (sAZ%GZ3Y ZAY*!Y' "Y*' Z1 Z>Y7$Z|j(j@r4|jA|jdz |jj|j(j,r|j-tC||j(j*r)|jE|j|jFz ||S)ahWrite inference results to a file or directory. Args: i (int): Index of the current image in the batch. p (Path): Path to the current image. im (torch.Tensor): Preprocessed image tensor. s (list[str]): List of result strings. Returns: (str): String with result information. rr>Nz: z frame (\d+)/r?rrbrxz {:g}x{:g} r@r[z.1fms) line_widthboxesrrim_gpuz.txt) save_confcrops)r file_name)$rcrHr)rqfrom_imgrXr&countresearchintrrYrr0rer-__str__rrrrr!plotr show_boxes show_conf show_labels retina_masksr(rrrstrsave_predicted_imagesname) r5rprPrstringframematchresults r9rzBasePredictor.write_resultss^ rxx=A DB    " "d&6&6&?&?4CSCSCZCZ 2h FLL&&EIIoqt4E%*CaME 0AFFDLLDUDUY`D`bhijoipfq4rs %,%%rxx|44a--//1V^^%&v||K'@&ERHH 99>>TYY^^%{{99//ii**YY((yy,,#yy55t2a5 + D  99   OOt}}oT2dii>Q>QO R 99     dmmg&=I[I[  \ 99>> IIc!f  99>>  & &t}}qvv'=u E r;c|j}|jjdvrt|jjdk(r|jjnd}|j|j dz }||j vr|jjrt|jddtrdn trdnd \}}tjtt|j!|tj"|||j$d |j$d f |j |<|j |j'||jjr*tj(|d |j d|d|yytj(t|j!d|y)zSave video predictions as mp4/avi or images as jpg at specified path. Args: save_path (Path): Path to save the results. frame (int): Frame number for video mode. >videorqr_framesTr)z.mp4avc1).aviWMV2)rMJPGr?r)filenamefourccfps frameSize/rxz.jpgN)r(r&rrrrYr'r save_framesrrTrrrrr with_suffixVideoWriter_fourccrHwriteimwrite)r5 save_pathrrPr frames_pathsuffixrs r9rz#BasePredictor.save_predicted_imagessn   <<   3 3&*ll&7&77&B$,,""C--Y^^,rs\!F#   12.2[[[>2^^ ++r;