|hAy PdZddlZddlZddlZddlZddlZddlZddlmZddl m Z m Z m Z m Z ddlZddlZddlmZmZddlmZmZddlmZddlmZmZmZmZmZm Z m!Z!m"Z"m#Z#dd l$m%Z%m&Z&m'Z'm(Z(m)Z)dd l*m+Z+dd l,m-Z-dd l.m/Z/m0Z0e"d z ddddddddf dZ1GddZ2GddZ3y)a Benchmark a YOLO model formats for speed and accuracy. Usage: from ultralytics.utils.benchmarks import ProfileModels, benchmark ProfileModels(['yolo11n.yaml', 'yolov8s.yaml']).run() benchmark(model='yolo11n.pt', imgsz=160) Format | `format=argument` | Model --- | --- | --- PyTorch | - | yolo11n.pt TorchScript | `torchscript` | yolo11n.torchscript ONNX | `onnx` | yolo11n.onnx OpenVINO | `openvino` | yolo11n_openvino_model/ TensorRT | `engine` | yolo11n.engine CoreML | `coreml` | yolo11n.mlpackage TensorFlow SavedModel | `saved_model` | yolo11n_saved_model/ TensorFlow GraphDef | `pb` | yolo11n.pb TensorFlow Lite | `tflite` | yolo11n.tflite TensorFlow Edge TPU | `edgetpu` | yolo11n_edgetpu.tflite TensorFlow.js | `tfjs` | yolo11n_web_model/ PaddlePaddle | `paddle` | yolo11n_paddle_model/ MNN | `mnn` | yolo11n.mnn NCNN | `ncnn` | yolo11n_ncnn_model/ IMX | `imx` | yolo11n_imx_model/ RKNN | `rknn` | yolo11n_rknn_model/ N)Path)ListOptionalTupleUnion)YOLO YOLOWorld) TASK2DATA TASK2METRIC)export_formats) ARM64ASSETS IS_JETSONLINUXLOGGERMACOSTQDM WEIGHTS_DIRYAML)IS_PYTHON_3_13 check_imgszcheck_requirements check_yolo is_rockchip) safe_download) file_size) get_cpu_info select_devicez yolo11n.ptFcpuMbP?c  L #$t|}t|tr|d|dk(sJd ddl$d$jj _d$jj _t|d}t|ttfr t|}t|jjd d d} |xst|j}t |j} g} t#j"} |j%}|r(t't)d }||vsJd |d |dt+t)j-D]T\}}}}}}d\}} |r||k7r|dk(r|jdk7st|t6rJd+| rJd,t.sJd-t=rJd.d/|j>vr |sJd0d1|j>vr |sJd2|d3k(r+|j@xs|jBxs |jD}|}nE|jFd`||||||dd4| }t||j5}|t|vsJd6d7}|jd8k7s |dk7sJd9|d:vsJd;|dk7stIjJd|jMtNd?z |||d@|jQ|d|d|||ddAB }|jR| |jTdC}}tWdD||zz dE}| jY|dFtWt[|dtW|dGtW|dE|gWte|J$jg| dKdLdM| dNdOgP}|jD}t#j"| z }dQ}dR|dS|dT|dU|dVdW|dX|jid3dX } tajj| tmdYdZd[d\]5}!|!jo| ddd|r@t|tpr0|| jr}"|#tu#$fd^|"Ds Jd_#|S#t\$rp}|rt?|t^us JdH|dI|tajbdH|dI|| jY||tWt[|ddddgYd}~d}~wwxYw#1swYxYw)aa Benchmark a YOLO model across different formats for speed and accuracy. Args: model (str | Path): Path to the model file or directory. data (str | None): Dataset to evaluate on, inherited from TASK2DATA if not passed. imgsz (int): Image size for the benchmark. half (bool): Use half-precision for the model if True. int8 (bool): Use int8-precision for the model if True. device (str): Device to run the benchmark on, either 'cpu' or 'cuda'. verbose (bool | float): If True or a float, assert benchmarks pass with given metric. eps (float): Epsilon value for divide by zero prevention. format (str): Export format for benchmarking. If not supplied all formats are benchmarked. **kwargs (Any): Additional keyword arguments for exporter. Returns: (pandas.DataFrame): A pandas DataFrame with benchmark results for each format, including file size, metric, and inference time. Examples: Benchmark a YOLO model with default settings: >>> from ultralytics.utils.benchmarks import benchmark >>> benchmark(model="yolo11n.pt", imgsz=640) rz'benchmark() only supports square imgsz.N xF)verboseend2endArgumentzExpected format to be one of z , but got 'z'.)❌Npbobbz.TensorFlow GraphDef not supported for OBB taskedgetpuz3Edge TPU export only supported on non-aarch64 Linux>tfjscoremlzECoreML and TF.js export only supported on macOS and non-aarch64 Linuxr0z#CoreML not supported on Python 3.13>r,r/tfliter. saved_modelz;YOLOWorldv2 TensorFlow exports not supported by onnx2tf yetpaddlez,YOLOWorldv2 Paddle exports not supported yetzBPaddle OBB bug https://github.com/PaddlePaddle/Paddle/issues/72024z3End-to-end models not supported by PaddlePaddle yetz3Windows and Jetson Paddle exports not supported yetmnnz)YOLOWorldv2 MNN exports not supported yetncnnz*YOLOWorldv2 NCNN exports not supported yetimxz%YOLOWorldv2 IMX exports not supporteddetectz%IMX only supported for detection taskC2fzIMX only supported for YOLOv8rknnz*YOLOWorldv2 RKNN exports not supported yetz+End-to-end models not supported by RKNN yetzRKNN only supported on Linuxz1RKNN Inference only supported on Rockchip devicesr zinference not supported on CPUcudazinference not supported on GPU-)imgszformathalfint8datadevicer')taskz export failedu❎posez(GraphDef Pose inference is not supported>r/r.zinference not supportedDarwinz(inference only supported on macOS>=10.13zHEnd-to-end torch.topk operation is not supported for NCNN prediction yetzbus.jpg)r<rAr>r'r!) r@batchr<plotsrAr>r?r'conf inferenceu✅zBenchmark failure for : )rAFormatu Status❔z Size (MB)zInference time (ms/im)FPS)columnsudBenchmarks legend: - ✅ Success - ❎ Export passed but validation failed - ❌️ Export failedz Benchmarks complete for z on z at imgsz=z (z.2fzs)  zbenchmarks.logaignoreutf-8)errorsencodingc3LK|]}j|s|kDywN)notna).0xfloorpds [/var/www/html/test/engine/venv/lib/python3.12/site-packages/ultralytics/utils/benchmarks.py zbenchmark..s=!1u9=s$ $z%Benchmark failure: metric(s) < floor );r isinstancelistpandasoptionsdisplay max_columnswidthrstrrrgetattrmodelr rBr timelower frozensetr zipvaluesrr rrr r__str__rtypept_path ckpt_path model_nameexportplatformsystempredictrval results_dictspeedroundappendr ExceptionAssertionErrorrerrorr DataFramefillnainfoopenwritefloatarrayall)%rir@r<r>r?rAr'epsr=kwargs is_end2endkeyyt0 format_argformatsnamesuffixr gpu_emojifilenameexported_modelresultsmetricrzfpsedfdtlegendsfmetricsr[r\s% @@r] benchmarkr4sH  E#-eT#:58uQx oFoo D%'BJJ""BJJ 65 1F%#t%U **2. 5AJ  (9UZZ(D ejj !C A BJN,Z89 d$A'+V`Uaac"dd -0.2B2I2I2K-LSU)ffc3%xQ UjF2~zzU*\,\\*9$Ua,aa*--u[5!)P+PP)KK%eY7v9vv7!%eY7g9gg7zzU*p,pp*%\'\\~iEp;pp9%eY7d9dd7%eY7e9ee7%%~%eY7`9``7zzX-V/VV- /P1PP/%eY7e9ee7%T'TT~<< e;e e9!44 O6O O4X%):h)F rHr rF%q'qq~  " "6I#5U6X\fk " l%(() G$005w}}[7QEF ,a0C HHdE58) HHdE58)>> benchmark = RF100Benchmark() >>> benchmark.set_key("your_roboflow_api_key") roboflowr)Roboflow)rN)rrrr)rrrs r]set_keyzRF100Benchmark.set_keys :&%7+r ds_link_txtctjjdr*tjdtj dfntj dtj dtj dtdt|d5}|D]} tjd|j\}}}}}|jj||d|} t| jsI|jj!|j#|j%|j'dnt)j*d |j,jtj.| z d z  d d d |j|j,fS#t0$rY)wxYw#1swY1xYw) a Parse dataset links and download datasets. Args: ds_link_txt (str): Path to the file containing dataset links. Returns: ds_names (List[str]): List of dataset names. ds_cfg_list (List[Path]): List of paths to dataset configuration files. Examples: >>> benchmark = RF100Benchmark() >>> benchmark.set_key("api_key") >>> benchmark.parse_dataset("datasets_links.txt") zrf-100zultralytics-benchmarkszQhttps://github.com/ultralytics/assets/releases/download/v0.0.0/datasets_links.txtrSrUz/+r;yolov8zDataset already downloaded.z data.yamlN)ospathexistsshutilrmtreemkdirchdirrrresplitstriprr|rr workspaceprojectversiondownloadrrrcwdr}) rrfilelinerurlrrr proj_versions r] parse_datasetzRF100Benchmark.parse_datasetsx :<9Qx "((8"45WYW_W_`hWi  )*ij + 0 D  :<((4:V7AsIwMM((1&-Yay#9L -446)))4<>> benchmark = RF100Benchmark() >>> benchmark.evaluate("path/to/data.yaml", "path/to/val_log.txt", "path/to/eval_log.txt", 0) )u🚀u⚠️u💡r+namesrSrc3&K|]}|v ywrWr_)rYsymbolrs r]r^z*RF100Benchmark.evaluate..FsA&v~As c |dk7S)Nr"r_)rxs r]z)RF100Benchmark.evaluate..Is #)rrPc 3~K|]4}|vs |dk(r)dvr%dvr!ddddddd d 6y w) rz(AP)z(AR)rr$rJrKrNr_)rYr class_namesentriess r]r^z*RF100Benchmark.evaluate..Ksl "K'AJ6;PU[cjUj")")!*#*1:%,QZ")!*!(!( "s:=Nr$zMultiple dicts foundrrrzSingle dict foundrrQrL)rrr readlinesanyrrafilterrextendlenrrrrr)rrrrr skip_symbolsrlines eval_linesrmap_vallstresrrrs @@@r]evaluatezRF100Benchmark.evaluate.s$9 ii *73 , 1 QKKMEJ ALAA**S/v&;WEF29:Q1774=:!! "% "    , z?Q  KK. /! +w<5(!'lG + KK+ ,/9:s7|:1=G -w 7 ?1 GGt}}X./r'"= > ?W~7;  <; ?W~s0AE7E2 E7 F:%F2E77FFN)zdatasets_links.txt) __name__ __module__ __qualname____doc__rrgrr staticmethodrrintrr_rr]rrsc$c,s,"#/#/J#t##7#7S77X[7rrceZdZdZ d deedededededed ed e e e jeffd Z d Zd ZedefdZed!dej&dedefdZd"dedefdZd"dedefdZdedeeefdeeefdeeeeeffdZededeeefdeeefdeeeeeffdZedeefdZy)# ProfileModelsa/ ProfileModels class for profiling different models on ONNX and TensorRT. This class profiles the performance of different models, returning results such as model speed and FLOPs. Attributes: paths (List[str]): Paths of the models to profile. num_timed_runs (int): Number of timed runs for the profiling. num_warmup_runs (int): Number of warmup runs before profiling. min_time (float): Minimum number of seconds to profile for. imgsz (int): Image size used in the models. half (bool): Flag to indicate whether to use FP16 half-precision for TensorRT profiling. trt (bool): Flag to indicate whether to profile using TensorRT. device (torch.device): Device used for profiling. Methods: run: Profile YOLO models for speed and accuracy across various formats. get_files: Get all relevant model files. get_onnx_model_info: Extract metadata from an ONNX model. iterative_sigma_clipping: Apply sigma clipping to remove outliers. profile_tensorrt_model: Profile a TensorRT model. profile_onnx_model: Profile an ONNX model. generate_table_row: Generate a table row with model metrics. generate_results_dict: Generate a dictionary of profiling results. print_table: Print a formatted table of results. Examples: Profile models and print results >>> from ultralytics.utils.benchmarks import ProfileModels >>> profiler = ProfileModels(["yolo11n.yaml", "yolov8s.yaml"], imgsz=640) >>> profiler.run() Npathsnum_timed_runsnum_warmup_runsmin_timer<r>trtrAc ||_||_||_||_||_||_||_t|tjr||_ yt||_ y)a] Initialize the ProfileModels class for profiling models. Args: paths (List[str]): List of paths of the models to be profiled. num_timed_runs (int): Number of timed runs for the profiling. num_warmup_runs (int): Number of warmup runs before the actual profiling starts. min_time (float): Minimum time in seconds for profiling a model. imgsz (int): Size of the image used during profiling. half (bool): Flag to indicate whether to use FP16 half-precision for TensorRT profiling. trt (bool): Flag to indicate whether to profile using TensorRT. device (torch.device | str | None): Device used for profiling. If None, it is determined automatically. Notes: FP16 'half' argument option removed for ONNX as slower on CPU than FP32. Examples: Initialize and profile models >>> from ultralytics.utils.benchmarks import ProfileModels >>> profiler = ProfileModels(["yolo11n.yaml", "yolov8s.yaml"], imgsz=640) >>> profiler.run() N) rrrrr<r>rr`torchrAr) rrrrrr<r>rrAs r]rzProfileModels.__init__sXB ,.     *65<< @f mTZF[ rc |j}|stjdgSg}g}|D]}|jd}|jdvrt t |}|j|j}|jr]|jjdk7rD|js4|jd|j|j|jd}|jd|j|jd }n%|jd k(r|j!|}|}n|j#t |} |j%t |} |j'|j)|j*| | ||j'|j-|j*| | ||j/||S) a Profile YOLO models for speed and accuracy across various formats including ONNX and TensorRT. Returns: (List[dict]): List of dictionaries containing profiling results for each model. Examples: Profile models and print results >>> from ultralytics.utils.benchmarks import ProfileModels >>> profiler = ProfileModels(["yolo11n.yaml", "yolov8s.yaml"]) >>> results = profiler.run() z'No matching *.pt or *.onnx files found.z.engine.pt.yml.yamlr engineF)r=r>r<rAr'onnx)r=r<rAr'z.onnx) get_filesrwarning with_suffixrrrgfuserrrArpis_filertr>r<get_onnx_model_infoprofile_tensorrt_modelprofile_onnx_modelr|generate_table_rowstemgenerate_results_dict print_table) rfiles table_rowsoutputr engine_fileri model_info onnx_filet_enginet_onnxs r]runzProfileModels.runs  NND EI  _D**95K{{66SY "ZZ\ 88 0 0E 9+BUBUBW"',,'!YY"jj#{{ % #/#K"LL!**;;! ) '!55d;  223{3CDH,,S^ $ rc >g}|jD]}t|}|jrLgd}|j|Dcgc]*}t jt ||z D]}|,c}}j|j dvr|jt ||jt jt |tjdt|t|Dcgc] }t|c}Scc}}wcc}w)z Return a list of paths for all relevant model files given by the user. Returns: (List[Path]): List of Path objects for the model files. )z*.ptz*.onnxz*.yamlrz Profiling: ) rris_dirrglobrgrr|rrsorted)rrr extensionsextrs r]r zProfileModels.get_filessJJ 3D:D{{}9  j`sTYYsSWZ]S]E_`Td`d`a 88 SY' TYYs4y12 3  k&-12'-e}5tT 55a6s /D ?Drcy)zWExtract metadata from an ONNX model file including parameters, GFLOPs, and input shape.)rrrrr_)rs r]rz!ProfileModels.get_onnx_model_infos"rr@sigma max_itersctj|}t|D]b}tj|tj|}}|||||zz kD||||zzkz}t |t |k(r|S|}d|S)a Apply iterative sigma clipping to data to remove outliers. Args: data (numpy.ndarray): Input data array. sigma (float): Number of standard deviations to use for clipping. max_iters (int): Maximum number of iterations for the clipping process. Returns: (numpy.ndarray): Clipped data array with outliers removed. )nprrangemeanstdr)r@r'r(rr,r- clipped_datas r]iterative_sigma_clippingz&ProfileModels.iterative_sigma_clippingsxx~y! A rvvd|#Dus{(:!:tdUUX[FX?X YZL< CI-  D   rrrcz|jrt|jsyt|}t j |j |j dftj}d}tdD]\}tj}t|jD]}|||j dtj|z }^tt|j||zz |jz|jdz}g} tt||D]8}|||j d} | j!| d j"d :|j%t j&| d d } t j(| t j*| fS) a Profile YOLO model performance with TensorRT, measuring average run time and standard deviation. Args: engine_file (str): Path to the TensorRT engine file. eps (float): Small epsilon value to prevent division by zero. Returns: mean_time (float): Mean inference time in milliseconds. std_time (float): Standard deviation of inference time in milliseconds. )rrr)dtyperF)r<r'2descrrHrJr'r()rrrrr*zerosr<uint8r+rjrmaxr{rrrr|rzr/rr,r-) rrrri input_dataelapsedr start_timenum_runs run_timesrs r]rz$ProfileModels.profile_tensorrt_modelsvxxtK088:[!XXtzz4::q9J q /AJ4//0 Cj EB CiikJ.G  /uT]]gm||zz |j6z|j@}g}tCt3||D]T}t5j4}|j9| g| | i|jEt5j4|z dzV|jGtjH|dd}tjJ|tjL|fS)a Profile an ONNX model, measuring average inference time and standard deviation across multiple runs. Args: onnx_file (str): Path to the ONNX model file. eps (float): Small epsilon value to prevent division by zero. Returns: mean_time (float): Mean inference time in milliseconds. std_time (float): Standard deviation of inference time in milliseconds. onnxruntimerNCPUExecutionProvider) providersc3JK|]}t|txr|dk\yw)rN)r`r)rYdims r]r^z3ProfileModels.profile_onnx_model..Ys#Z*S#.;3!8;Zs!#r$rfloat16rdoubleint64int32zUnsupported ONNX datatype rr3rIrJrr5)'rr?SessionOptionsGraphOptimizationLevelORT_ENABLE_ALLgraph_optimization_levelintra_op_num_threadsInferenceSession get_inputsrprshaper<r*rEfloat32float64rGrH ValueErrorrandomrandastyper get_outputsr+rjrrr8r{rrrr|r/rr,r-)rrrort sess_optionssess input_tensor input_typedynamic input_shape input_dtyper9 input_name output_namer:rr;r<r=s r]rz ProfileModels.profile_onnx_modelBs =)!))+ 030J0J0Y0Y -,- )##I|H^G_#`(+ !&& Z|GYGYZZZ8?q!TZZ4\EWEW   "**K  "**K  #**K  "((K  "((K9*FG GYY^^[188E !&& &&(+00 q /AJ4//0 B+Z(@A BiikJ.G  /uT]]gm ? @ 11"((92EQZ[1\ wwy!266)#444rrsrrrc|\}}}}d|dd|jd|ddd|ddd |ddd|ddd |d z dd|dd S) a Generate a table row string with model performance metrics. Args: model_name (str): Name of the model. t_onnx (tuple): ONNX model inference time statistics (mean, std). t_engine (tuple): TensorRT engine inference time statistics (mean, std). model_info (tuple): Model information (layers, params, gradients, flops). Returns: (str): Formatted table row string with model metrics. z| 18sz | z | - | rz.1f±r$z ms | g.Az |)r<) rrsrrrlayersparams gradientsflopss r]rz ProfileModels.generate_table_rows&,6( 5C DJJ(pixels)zmAPval
50-95zSpeed
CPU (z) ONNX
(ms)zSpeed
z TensorRT
(ms)zparams
(M)zFLOPs
(B)|c3(K|] }d|d yw)rNr_rYhs r]r^z,ProfileModels.print_table..s:Q!A3a:sc3>K|]}dt|dzzyw)r;rJN)rros r]r^z,ProfileModels.print_table..s"G!3#a&1*#5"Gsz N)rr: is_availableget_device_namerjoinrr)rrheadersheader separatorrows r]rzProfileModels.print_tables05zz/F/F/Hejj((+e  # "!,.!1 @SE!2 3  sxx:':::S@#(("Gw"GGG#M  d6(O$ I C KK  r)dr%<iTTN)rJr)r!)rrrrrrgrrboolrrrrArrr rrr*ndarrayr/rrrrrrr_rr]rrhsH"!59(\Cy(\(\ (\  (\  (\(\(\u||S012(\T5n6*"s""rzz%PS*%5#%5E%5N?5C?5e?5B  eUl# u %  %u45  2  eUl# u % %u45   6S rr)4rr"rrurrrjpathlibrtypingrrrrnumpyr* torch.cudar ultralyticsrr ultralytics.cfgr r ultralytics.engine.exporterr ultralytics.utilsr rrrrrrrrultralytics.utils.checksrrrrrultralytics.utils.downloadsrultralytics.utils.filesrultralytics.utils.torch_utilsrrrrrr_rr]rs8  //'26eeemm5-E  $     _DOOdhhr