.hv} HdZddlmZddlZddlZddlZddlZddlZddlZddl m Z ddl Z ddl ZddlmZmZddlmZmZddlmZddlmZmZmZmZmZmZmZmZm 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-m.Z.e d z ddddddddf dZ/GddZ0GddZ1y)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/ ) annotationsN)Path)YOLO YOLOWorld) TASK2DATA TASK2METRIC)export_formats) ARM64ASSETS ASSETS_URL 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  %t|}t|tr|d|dk(sJd ddl} | jj d| jj d| jjd| jjd| jjd| jjdt|d }t|ttfr t|}t|j j dd d} |xst"|j$}t&|j$} g} t)j(}|j+}|r(t-t/d }||vsJd |d |dt1t/j3D]T\}}}}}}d\}} |r||k7r|dk(r|j$dk7sr t8sJd|d k(rt|t<rJd!|d"k(rt|t<rJd#|d$k(rJ| rJt|t<rJd%|j$d&k(sJd'd(|jAvsJd)|d*k(r>t|t<rJd+| rJd,t4sJd-tCrJd.d/|jDvr |sJd0d1|jDvr |sJd2|d3k(r+|jFxs|jHxs |jJ}|}nE|jLdc||||||dd4| }t||j$5}|t|vsJd6d7}|j$d8k7s |dk7sJd9|d:vsJd;|dk7stOjPd|jStTd?z |||d@|jW|d|d|||ddAB }|jX| |jZdC}}t]dD||zz dE}| j_|dFt]ta|dt]|dGt]|dE|gWtk|J| jm| dKdLdM| dNdOgdPQ}|jodRdS}|jq| jsju| jvjyd3}|jJ}t)j(|z } dT}!dU|dV|dW|dX| dYdZ|!d[|d[ }"tgjz|"t}d\d]d^d_`5}#|#j|"ddd|rCt|tr3|| j}$|%ts%fda|$Ds Jdb%|S#tb$rp}|rtE|tdus JdH|dI|tgjhdH|dI|| j_||t]ta|ddddgYd}~ d}~wwxYw#1swYxYw)da 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: (polars.DataFrame): A polars 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) rTz'benchmark() only supports square imgsz.NASCII_BORDERS_ONLY_CONDENSEDF)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 Linuxr-z#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 taskC2fz*IMX only supported for YOLOv8n and YOLO11nrknnz*YOLOWorldv2 RKNN exports not supported yetz+End-to-end models not supported by RKNN yetzRKNN only supported on Linuxz1RKNN Inference only supported on Rockchip devicesrzinference 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)r9r>r;r%r) r=batchr9plotsr>r;r<r%conf inferenceu✅zBenchmark failure for : )r>Formatu Status❔z Size (MB)zInference time (ms/im)FPSrow)schemaorient )offsetudBenchmarks legend: - ✅ Success - ❎ Export passed but validation failed - ❌️ Export failedz Benchmarks complete for z on z at imgsz=z (z.2fzs)  zbenchmarks.logaignoreutf-8)errorsencodingc3TK|]}tj|r|kD!ywN)npisnan).0xfloors Z/var/www/html/ai-service/venv/lib/python3.12/site-packages/ultralytics/utils/benchmarks.py zbenchmark..sARXXa[1u9As( (z%Benchmark failure: metric(s) < floor )Br isinstancelistpolarsConfig set_tbl_cols set_tbl_rowsset_tbl_width_charsset_tbl_hide_column_data_typesset_tbl_hide_dataframe_shapeset_tbl_formattingrstrrrgetattrmodelrr?rtimelower frozensetr zipvaluesrr rrrr __str__rtypept_path ckpt_path model_nameexportplatformsystempredictr val results_dictspeedroundappendr ExceptionAssertionErrorrerrorr DataFramewith_row_index with_columnsallcastString fill_nullinfoopenwritefloatto_numpy)&rmr=r9r;r<r>r%epsr:kwargspl is_end2endkeyyt0 format_argformatsnamesuffixrgpu_emojifilenameexported_modelresultsmetricr~fpsedf df_displaydtlegendsfmetricsr]s& @r^ benchmarkr5s(H  E#-eT#:58uQx oFoo DII2II2II!!"%II,,T2II**40II  !?@ 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- /]1]]/%eY7e9ee7%T'TT~<< e;e e9!44 O6O O4X%):h)F rHr rF%q'qq~  " "6I#5U6X\fk " l%(() G$005w}}[7QEF ,a0C HHdE58) HHdE58)>> benchmark = RF100Benchmark() >>> benchmark.set_key("your_roboflow_api_key") roboflowr)Roboflow)api_keyN)rrrr)rrrs r^set_keyzRF100Benchmark.set_keys :&%7+rctjjdr*tjdtj dfntj dtj dtj dttdt|d5}|D]} tjd|j\}}}}}|jj||d|} t| jsI|j j#|j%|j'|j)dnt+j,d |j.jtj0| z d z  d d d |j|j.fS#t2$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-benchmarksz/datasets_links.txtrTrVz/+r8yolov8zDataset already downloaded.z data.yamlN)ospathexistsshutilrmtreemkdirchdirrr rresplitstriprrrr workspaceprojectversiondownloadrrrcwdr) r ds_link_txtfilelinerurlrrr proj_versions r^ parse_datasetzRF100Benchmark.parse_dataset s :<9Qx "((8"45WYW_W_`hWi  )* $789 + 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(namesrTrc3&K|]}|v ywrXr`)r[symbolrs r^r_z*RF100Benchmark.evaluate..NsA&v~AsrOc |dk7S)Nr r`)r|s r^z)RF100Benchmark.evaluate..Qs #)rrQc 3~K|]4}|vs |dk(r)dvr%dvr!ddddddd d 6y w) rz(AP)z(AR)rr"rGrHrNr`)r[r class_namesentriess r^r_z*RF100Benchmark.evaluate..Ssl "K'AJ6;PU[cjUj")")!*#*1:%,QZ")!*!(!( "s:=Nr"zMultiple dicts foundrrrzSingle dict foundrrRrI)rrr readlinesanyrrbfilterrextendlenrrrrr)r yaml_path val_log_file eval_log_filelist_ind skip_symbolsrlines eval_linesrmap_vallstresrrrs @@@r^evaluatezRF100Benchmark.evaluate6s$9 ii *73 , 1 QKKMEJ ALAA**S/v&;WEF29:Q1774=:!! "% "    , z?Q  KK. /! +w<5(!'lG + KK+ ,/9:s7|:1=G -w 7 ?1 GGt}}X./r'"= > ?W~7;  <; ?W~s0AE7E2 E7 F:%F2E77FFN)rrk)zdatasets_links.txt)rrk)rr)rrkrrkrrkrint) __name__ __module__ __qualname____doc__rrr staticmethodrrr`rr^rrs1$c,"#/J##7rrceZdZdZ d ddZdZdZeddZedddZ dddZ ed Z ddd Z dd Z e dd Zedd Zy) 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() Nc ||_||_||_||_||_||_||_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) pathsnum_timed_runsnum_warmup_runsmin_timer9r;trtratorchr>r) rrrr r r9r;r r>s 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.yamlrengineF)r:r;r9r>r%onnx)r:r9r>r%z.onnx) get_filesrwarning with_suffixrrrkfuserr r>rtis_filerxr;r9get_onnx_model_infoprofile_tensorrt_modelprofile_onnx_modelrgenerate_table_rowstemgenerate_results_dict print_table) rfiles table_rowsoutputr engine_filerm 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_dirrglobrkrrrrsorted)rr r extensionsextrs r^rzProfileModels.get_filessJJ 3D:D{{}9  j`sTYYsSWZ]S]E_`Td`d`a 88 SY' TYYs4y12 3  k&-12'-e}5tT 55a6s /D ?Dcy)zWExtract metadata from an ONNX model file including parameters, GFLOPs, and input shape.)rrrrr`)r%s r^rz!ProfileModels.get_onnx_model_infos"rctj|}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 (np.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: (np.ndarray): Clipped data array with outliers removed. )rYarrayrangemeanstdr)r=sigma max_itersrr3r4 clipped_datas r^iterative_sigma_clippingz&ProfileModels.iterative_sigma_clipping sxx~y! A rvvd|#Dus{(:!:tdUUX[FX?X YZL< CI-  D   rcz|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)r9r%2descrrErGr5r6)r rrrrYzerosr9uint8r2rnr maxrr rrrr~r8r1r3r4) rr#rrm input_dataelapsedr start_timenum_runs run_timesrs r^rz$ProfileModels.profile_tensorrt_model#svxxtK088:[!XXtzz4::q9J q /AJ4//0 Cj EB CiikJ.G  /uT]]gm.Ms#QSz#s+8q8Qs!#)r) tensor_shapes r^ check_dynamiczProfileModels.check_dynamicJsQLQQQQrctdgddl}|j}|jj|_d|_|j||dg}t}|jD]}|j}|j|jrt|jdk7rC|j|jddr%td |jd |jt|jdk(rdd |j |j fndg|jdd} n |j} d |vrt"j$} nbd |vrt"j&} nMd|vrt"j(} n8d|vrt"j*} n#d|vrt"j,} ntd|t#j.j0| j3| } |j} |j5| | i|j7dj} d}t9d D]Z}t;j:}t9|j<D]}|j?| g|t;j:|z }\tAtC|jD||zz |j<z|jF}g}tIt9||D]R}t;j:}|j?| g||jKt;j:|z dzT|jMt#jN|dd}t#jP|t#jR|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. ) onnxruntimezonnxruntime-gpurNCPUExecutionProvider) providersrHr"zUnsupported dynamic shape z of rfloat16rdoubleint64int32zUnsupported ONNX datatype rr<rFrGrr>)*rrMSessionOptionsGraphOptimizationLevelORT_ENABLE_ALLgraph_optimization_levelintra_op_num_threadsInferenceSessiondict get_inputsrtrKshaper ValueErrorrr9rYrQfloat32float64rSrTrandomrandastypeupdate get_outputsr2rnr r(rArr rrrr8r1r3r4)rr%rort sess_optionssessinput_data_dict input_tensor input_type input_shape input_dtyperB input_name output_namerCrrDrErFs r^rz ProfileModels.profile_onnx_modelOs/ >?@!))+ 030J0J0Y0Y -,- )##I|H^G_#`& OO- =L%**J!!,"4"45|))*a/D4F4F|GYGYZ[Z\G]4^$'A,BTBTAUUYZfZkZkYl%mnn69,:L:L6MQR6RQ4::tzz2YZXt]i]o]opqpr]sXt+00 J& jj J& jj Z' jj J& hh J& hh  #=j\!JKK5<<[IJ%**J  " "J #; <7 =:&&(+00 q /AJ4//0 9+8 9iikJ.G  /uT]]gm ? @ 11"((92EQZ[1\ wwy!266)#444rc|\}}}}d|dd|jd|ddd|ddd |ddd|ddd |d z dd|dd S) a Generate a table row string with model performance metrics. Args: model_name (str): Name of the model. t_onnx (tuple): ONNX model inference time statistics (mean, std). t_engine (tuple): TensorRT engine inference time statistics (mean, std). model_info (tuple): Model information (layers, params, gradients, flops). Returns: (str): Formatted table row string with model metrics. z| 18sz | z | - | rz.1f±r"z ms | g.Az |)r9) rrwr'r&r$layersparams 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)rONr`r[hs r^r_z,ProfileModels.print_table..s:Q!A3a:sc3>K|]}dt|dzzyw)r8rGN)rr}s r^r_z,ProfileModels.print_table..s"G!3#a&1*#5"Gsz N)r r7 is_availableget_device_namerjoinrr)r!rheadersheader separatorrLs r^rzProfileModels.print_tables05zz/F/F/Hejj((+e  # "!,.!1 @SE!2 3  sxx:':::S@#(("Gw"GGG#M  d6(O$ I C KK  r)d <iTTN)r list[str]rrr rr rr9rr;boolr rr>ztorch.device | str | None)r%rk)rGr)r=z np.ndarrayr5rr6r)r)r#rkrr)r%rkrr)rwrkr'tuple[float, float]r&rr$z!tuple[float, float, float, float])r!r)rrrrrr(rrrr8rrKrrrrr`rr^rrpsIH"!,0(\(\(\ (\  (\  (\(\(\*(\T5n6*""*%5NRRH5T  $ &  6  2  # & 6   6rr)2r __future__rr+rryrrrnpathlibrnumpyrY torch.cudar  ultralyticsrrultralytics.cfgrrultralytics.engine.exporterr ultralytics.utilsr r r r rrrrrrultralytics.utils.checksrrrrrultralytics.utils.downloadsrultralytics.utils.filesrultralytics.utils.torch_utilsrrrrrr`rr^rs8#  '26qqqmm5-E  $     fROOdvvr