انتقل إلى المحتوى

مرجع ل ultralytics/utils/benchmarks.py

ملاحظه

هذا الملف متاح في https://github.com/ultralytics/ultralytics/ نقطة / الرئيسية /ultralytics/ المرافق / المعايير.py. إذا اكتشفت مشكلة ، فيرجى المساعدة في إصلاحها من خلال المساهمة في طلب 🛠️ سحب. شكرا لك 🙏!



ultralytics.utils.benchmarks.ProfileModels

فئة ProfileModels لتنميط نماذج مختلفة على ONNX و TensorRT.

تحدد هذه الفئة أداء الطرز المختلفة ، وتعيد النتائج مثل سرعة النموذج و FLOPs.

سمات:

اسم نوع وصف
paths list

مسارات النماذج إلى الملف الشخصي.

num_timed_runs int

عدد عمليات التشغيل الموقوتة للتنميط. الافتراضي هو 100.

num_warmup_runs int

عدد مرات الإحماء قبل التنميط. الافتراضي هو 10.

min_time float

الحد الأدنى لعدد الثواني للملف الشخصي. الافتراضي هو 60.

imgsz int

حجم الصورة المستخدمة في النماذج. الافتراضي هو 640.

أساليب:

اسم وصف
profile

ملامح النماذج وطباعة النتيجة.

مثل
from ultralytics.utils.benchmarks import ProfileModels

ProfileModels(['yolov8n.yaml', 'yolov8s.yaml'], imgsz=640).profile()
شفرة المصدر في ultralytics/utils/benchmarks.py
class ProfileModels:
    """
    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): Paths of the models to profile.
        num_timed_runs (int): Number of timed runs for the profiling. Default is 100.
        num_warmup_runs (int): Number of warmup runs before profiling. Default is 10.
        min_time (float): Minimum number of seconds to profile for. Default is 60.
        imgsz (int): Image size used in the models. Default is 640.

    Methods:
        profile(): Profiles the models and prints the result.

    Example:
        ```python
        from ultralytics.utils.benchmarks import ProfileModels

        ProfileModels(['yolov8n.yaml', 'yolov8s.yaml'], imgsz=640).profile()
        ```
    """

    def __init__(
        self,
        paths: list,
        num_timed_runs=100,
        num_warmup_runs=10,
        min_time=60,
        imgsz=640,
        half=True,
        trt=True,
        device=None,
    ):
        """
        Initialize the ProfileModels class for profiling models.

        Args:
            paths (list): List of paths of the models to be profiled.
            num_timed_runs (int, optional): Number of timed runs for the profiling. Default is 100.
            num_warmup_runs (int, optional): Number of warmup runs before the actual profiling starts. Default is 10.
            min_time (float, optional): Minimum time in seconds for profiling a model. Default is 60.
            imgsz (int, optional): Size of the image used during profiling. Default is 640.
            half (bool, optional): Flag to indicate whether to use half-precision floating point for profiling.
            trt (bool, optional): Flag to indicate whether to profile using TensorRT. Default is True.
            device (torch.device, optional): Device used for profiling. If None, it is determined automatically.
        """
        self.paths = paths
        self.num_timed_runs = num_timed_runs
        self.num_warmup_runs = num_warmup_runs
        self.min_time = min_time
        self.imgsz = imgsz
        self.half = half
        self.trt = trt  # run TensorRT profiling
        self.device = device or torch.device(0 if torch.cuda.is_available() else "cpu")

    def profile(self):
        """Logs the benchmarking results of a model, checks metrics against floor and returns the results."""
        files = self.get_files()

        if not files:
            print("No matching *.pt or *.onnx files found.")
            return

        table_rows = []
        output = []
        for file in files:
            engine_file = file.with_suffix(".engine")
            if file.suffix in (".pt", ".yaml", ".yml"):
                model = YOLO(str(file))
                model.fuse()  # to report correct params and GFLOPs in model.info()
                model_info = model.info()
                if self.trt and self.device.type != "cpu" and not engine_file.is_file():
                    engine_file = model.export(
                        format="engine", half=self.half, imgsz=self.imgsz, device=self.device, verbose=False
                    )
                onnx_file = model.export(
                    format="onnx", half=self.half, imgsz=self.imgsz, simplify=True, device=self.device, verbose=False
                )
            elif file.suffix == ".onnx":
                model_info = self.get_onnx_model_info(file)
                onnx_file = file
            else:
                continue

            t_engine = self.profile_tensorrt_model(str(engine_file))
            t_onnx = self.profile_onnx_model(str(onnx_file))
            table_rows.append(self.generate_table_row(file.stem, t_onnx, t_engine, model_info))
            output.append(self.generate_results_dict(file.stem, t_onnx, t_engine, model_info))

        self.print_table(table_rows)
        return output

    def get_files(self):
        """Returns a list of paths for all relevant model files given by the user."""
        files = []
        for path in self.paths:
            path = Path(path)
            if path.is_dir():
                extensions = ["*.pt", "*.onnx", "*.yaml"]
                files.extend([file for ext in extensions for file in glob.glob(str(path / ext))])
            elif path.suffix in {".pt", ".yaml", ".yml"}:  # add non-existing
                files.append(str(path))
            else:
                files.extend(glob.glob(str(path)))

        print(f"Profiling: {sorted(files)}")
        return [Path(file) for file in sorted(files)]

    def get_onnx_model_info(self, onnx_file: str):
        """Retrieves the information including number of layers, parameters, gradients and FLOPs for an ONNX model
        file.
        """
        return 0.0, 0.0, 0.0, 0.0  # return (num_layers, num_params, num_gradients, num_flops)

    def iterative_sigma_clipping(self, data, sigma=2, max_iters=3):
        """Applies an iterative sigma clipping algorithm to the given data times number of iterations."""
        data = np.array(data)
        for _ in range(max_iters):
            mean, std = np.mean(data), np.std(data)
            clipped_data = data[(data > mean - sigma * std) & (data < mean + sigma * std)]
            if len(clipped_data) == len(data):
                break
            data = clipped_data
        return data

    def profile_tensorrt_model(self, engine_file: str, eps: float = 1e-3):
        """Profiles the TensorRT model, measuring average run time and standard deviation among runs."""
        if not self.trt or not Path(engine_file).is_file():
            return 0.0, 0.0

        # Model and input
        model = YOLO(engine_file)
        input_data = np.random.rand(self.imgsz, self.imgsz, 3).astype(np.float32)  # must be FP32

        # Warmup runs
        elapsed = 0.0
        for _ in range(3):
            start_time = time.time()
            for _ in range(self.num_warmup_runs):
                model(input_data, imgsz=self.imgsz, verbose=False)
            elapsed = time.time() - start_time

        # Compute number of runs as higher of min_time or num_timed_runs
        num_runs = max(round(self.min_time / (elapsed + eps) * self.num_warmup_runs), self.num_timed_runs * 50)

        # Timed runs
        run_times = []
        for _ in TQDM(range(num_runs), desc=engine_file):
            results = model(input_data, imgsz=self.imgsz, verbose=False)
            run_times.append(results[0].speed["inference"])  # Convert to milliseconds

        run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=3)  # sigma clipping
        return np.mean(run_times), np.std(run_times)

    def profile_onnx_model(self, onnx_file: str, eps: float = 1e-3):
        """Profiles an ONNX model by executing it multiple times and returns the mean and standard deviation of run
        times.
        """
        check_requirements("onnxruntime")
        import onnxruntime as ort

        # Session with either 'TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'
        sess_options = ort.SessionOptions()
        sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
        sess_options.intra_op_num_threads = 8  # Limit the number of threads
        sess = ort.InferenceSession(onnx_file, sess_options, providers=["CPUExecutionProvider"])

        input_tensor = sess.get_inputs()[0]
        input_type = input_tensor.type

        # Mapping ONNX datatype to numpy datatype
        if "float16" in input_type:
            input_dtype = np.float16
        elif "float" in input_type:
            input_dtype = np.float32
        elif "double" in input_type:
            input_dtype = np.float64
        elif "int64" in input_type:
            input_dtype = np.int64
        elif "int32" in input_type:
            input_dtype = np.int32
        else:
            raise ValueError(f"Unsupported ONNX datatype {input_type}")

        input_data = np.random.rand(*input_tensor.shape).astype(input_dtype)
        input_name = input_tensor.name
        output_name = sess.get_outputs()[0].name

        # Warmup runs
        elapsed = 0.0
        for _ in range(3):
            start_time = time.time()
            for _ in range(self.num_warmup_runs):
                sess.run([output_name], {input_name: input_data})
            elapsed = time.time() - start_time

        # Compute number of runs as higher of min_time or num_timed_runs
        num_runs = max(round(self.min_time / (elapsed + eps) * self.num_warmup_runs), self.num_timed_runs)

        # Timed runs
        run_times = []
        for _ in TQDM(range(num_runs), desc=onnx_file):
            start_time = time.time()
            sess.run([output_name], {input_name: input_data})
            run_times.append((time.time() - start_time) * 1000)  # Convert to milliseconds

        run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=5)  # sigma clipping
        return np.mean(run_times), np.std(run_times)

    def generate_table_row(self, model_name, t_onnx, t_engine, model_info):
        """Generates a formatted string for a table row that includes model performance and metric details."""
        layers, params, gradients, flops = model_info
        return f"| {model_name:18s} | {self.imgsz} | - | {t_onnx[0]:.2f} ± {t_onnx[1]:.2f} ms | {t_engine[0]:.2f} ± {t_engine[1]:.2f} ms | {params / 1e6:.1f} | {flops:.1f} |"

    def generate_results_dict(self, model_name, t_onnx, t_engine, model_info):
        """Generates a dictionary of model details including name, parameters, GFLOPS and speed metrics."""
        layers, params, gradients, flops = model_info
        return {
            "model/name": model_name,
            "model/parameters": params,
            "model/GFLOPs": round(flops, 3),
            "model/speed_ONNX(ms)": round(t_onnx[0], 3),
            "model/speed_TensorRT(ms)": round(t_engine[0], 3),
        }

    def print_table(self, table_rows):
        """Formats and prints a comparison table for different models with given statistics and performance data."""
        gpu = torch.cuda.get_device_name(0) if torch.cuda.is_available() else "GPU"
        header = f"| Model | size<br><sup>(pixels) | mAP<sup>val<br>50-95 | Speed<br><sup>CPU ONNX<br>(ms) | Speed<br><sup>{gpu} TensorRT<br>(ms) | params<br><sup>(M) | FLOPs<br><sup>(B) |"
        separator = "|-------------|---------------------|--------------------|------------------------------|-----------------------------------|------------------|-----------------|"

        print(f"\n\n{header}")
        print(separator)
        for row in table_rows:
            print(row)

__init__(paths, num_timed_runs=100, num_warmup_runs=10, min_time=60, imgsz=640, half=True, trt=True, device=None)

تهيئة الفئة ProfileModels لنماذج التنميط.

البارامترات:

اسم نوع وصف افتراضي
paths list

قائمة مسارات النماذج التي سيتم توصيفها.

مطلوب
num_timed_runs int

عدد عمليات التشغيل الموقوتة للتنميط. الافتراضي هو 100.

100
num_warmup_runs int

عدد عمليات الإحماء قبل بدء التنميط الفعلي. الافتراضي هو 10.

10
min_time float

الحد الأدنى من الوقت بالثواني لتنميط نموذج. الافتراضي هو 60.

60
imgsz int

حجم الصورة المستخدمة أثناء التنميط. الافتراضي هو 640.

640
half bool

ضع علامة للإشارة إلى ما إذا كان سيتم استخدام نقطة عائمة نصف دقيقة للتنميط.

True
trt bool

وضع علامة للإشارة إلى ما إذا كنت تريد استخدام ملف التعريف أم لا TensorRT. الافتراضي هو صواب.

True
device device

الجهاز المستخدم للتنميط. إذا لم يكن هناك شيء ، يتم تحديده تلقائيا.

None
شفرة المصدر في ultralytics/utils/benchmarks.py
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188189 190 191 192 193 194 195 196 197 198 199 200 201 202203
def __init__(
    self,
    paths: list,
    num_timed_runs=100,
    num_warmup_runs=10,
    min_time=60,
    imgsz=640,
    half=True,
    trt=True,
    device=None,
):
    """
    Initialize the ProfileModels class for profiling models.

    Args:
        paths (list): List of paths of the models to be profiled.
        num_timed_runs (int, optional): Number of timed runs for the profiling. Default is 100.
        num_warmup_runs (int, optional): Number of warmup runs before the actual profiling starts. Default is 10.
        min_time (float, optional): Minimum time in seconds for profiling a model. Default is 60.
        imgsz (int, optional): Size of the image used during profiling. Default is 640.
        half (bool, optional): Flag to indicate whether to use half-precision floating point for profiling.
        trt (bool, optional): Flag to indicate whether to profile using TensorRT. Default is True.
        device (torch.device, optional): Device used for profiling. If None, it is determined automatically.
    """
    self.paths = paths
    self.num_timed_runs = num_timed_runs
    self.num_warmup_runs = num_warmup_runs
    self.min_time = min_time
    self.imgsz = imgsz
    self.half = half
    self.trt = trt  # run TensorRT profiling
    self.device = device or torch.device(0 if torch.cuda.is_available() else "cpu")

generate_results_dict(model_name, t_onnx, t_engine, model_info)

ينشئ قاموسا لتفاصيل النموذج بما في ذلك الاسم والمعلمات و GFLOPS ومقاييس السرعة.

شفرة المصدر في ultralytics/utils/benchmarks.py
364 365 366 367 368369 370 371 372 373
def generate_results_dict(self, model_name, t_onnx, t_engine, model_info):
    """Generates a dictionary of model details including name, parameters, GFLOPS and speed metrics."""
    layers, params, gradients, flops = model_info
    return {
        "model/name": model_name,
        "model/parameters": params,
        "model/GFLOPs": round(flops, 3),
        "model/speed_ONNX(ms)": round(t_onnx[0], 3),
        "model/speed_TensorRT(ms)": round(t_engine[0], 3),
    }

generate_table_row(model_name, t_onnx, t_engine, model_info)

ينشئ سلسلة منسقة لصف جدول يتضمن أداء النموذج وتفاصيل القياس.

شفرة المصدر في ultralytics/utils/benchmarks.py
def generate_table_row(self, model_name, t_onnx, t_engine, model_info):
    """Generates a formatted string for a table row that includes model performance and metric details."""
    layers, params, gradients, flops = model_info
    return f"| {model_name:18s} | {self.imgsz} | - | {t_onnx[0]:.2f} ± {t_onnx[1]:.2f} ms | {t_engine[0]:.2f} ± {t_engine[1]:.2f} ms | {params / 1e6:.1f} | {flops:.1f} |"

get_files()

ترجع قائمة مسارات لكل ملفات النموذج ذات الصلة التي قدمها المستخدم.

شفرة المصدر في ultralytics/utils/benchmarks.py
242 243 244 245 246 247 248249 250 251 252 253 254 255 256
def get_files(self):
    """Returns a list of paths for all relevant model files given by the user."""
    files = []
    for path in self.paths:
        path = Path(path)
        if path.is_dir():
            extensions = ["*.pt", "*.onnx", "*.yaml"]
            files.extend([file for ext in extensions for file in glob.glob(str(path / ext))])
        elif path.suffix in {".pt", ".yaml", ".yml"}:  # add non-existing
            files.append(str(path))
        else:
            files.extend(glob.glob(str(path)))

    print(f"Profiling: {sorted(files)}")
    return [Path(file) for file in sorted(files)]

get_onnx_model_info(onnx_file)

يسترجع المعلومات بما في ذلك عدد الطبقات والمعلمات والتدرجات و FLOPs ل ONNX نموذج ملف.

شفرة المصدر في ultralytics/utils/benchmarks.py
def get_onnx_model_info(self, onnx_file: str):
    """Retrieves the information including number of layers, parameters, gradients and FLOPs for an ONNX model
    file.
    """
    return 0.0, 0.0, 0.0, 0.0  # return (num_layers, num_params, num_gradients, num_flops)

iterative_sigma_clipping(data, sigma=2, max_iters=3)

يطبق خوارزمية قص سيغما تكرارية على البيانات المحددة مضروبة في عدد التكرارات.

شفرة المصدر في ultralytics/utils/benchmarks.py
264 265 266 267 268269 270 271 272 273
def iterative_sigma_clipping(self, data, sigma=2, max_iters=3):
    """Applies an iterative sigma clipping algorithm to the given data times number of iterations."""
    data = np.array(data)
    for _ in range(max_iters):
        mean, std = np.mean(data), np.std(data)
        clipped_data = data[(data > mean - sigma * std) & (data < mean + sigma * std)]
        if len(clipped_data) == len(data):
            break
        data = clipped_data
    return data

print_table(table_rows)

تنسيق وطباعة جدول مقارنة لنماذج مختلفة مع إحصائيات وبيانات أداء معينة.

شفرة المصدر في ultralytics/utils/benchmarks.py
375 376 377 378 379 380 381 382 383 384
def print_table(self, table_rows):
    """Formats and prints a comparison table for different models with given statistics and performance data."""
    gpu = torch.cuda.get_device_name(0) if torch.cuda.is_available() else "GPU"
    header = f"| Model | size<br><sup>(pixels) | mAP<sup>val<br>50-95 | Speed<br><sup>CPU ONNX<br>(ms) | Speed<br><sup>{gpu} TensorRT<br>(ms) | params<br><sup>(M) | FLOPs<br><sup>(B) |"
    separator = "|-------------|---------------------|--------------------|------------------------------|-----------------------------------|------------------|-----------------|"

    print(f"\n\n{header}")
    print(separator)
    for row in table_rows:
        print(row)

profile()

يسجل نتائج قياس الأداء لنموذج ما، ويتحقق من المقاييس مقابل الحد الأدنى ويعيد النتائج.

شفرة المصدر في ultralytics/utils/benchmarks.py
205 206 207 208 209 210 211 212 213 214 215 216 217 218219 220 221 222 223 224 225 226 227228229 230 231 232 233 234235236 237238 239240
def profile(self):
    """Logs the benchmarking results of a model, checks metrics against floor and returns the results."""
    files = self.get_files()

    if not files:
        print("No matching *.pt or *.onnx files found.")
        return

    table_rows = []
    output = []
    for file in files:
        engine_file = file.with_suffix(".engine")
        if file.suffix in (".pt", ".yaml", ".yml"):
            model = YOLO(str(file))
            model.fuse()  # to report correct params and GFLOPs in model.info()
            model_info = model.info()
            if self.trt and self.device.type != "cpu" and not engine_file.is_file():
                engine_file = model.export(
                    format="engine", half=self.half, imgsz=self.imgsz, device=self.device, verbose=False
                )
            onnx_file = model.export(
                format="onnx", half=self.half, imgsz=self.imgsz, simplify=True, device=self.device, verbose=False
            )
        elif file.suffix == ".onnx":
            model_info = self.get_onnx_model_info(file)
            onnx_file = file
        else:
            continue

        t_engine = self.profile_tensorrt_model(str(engine_file))
        t_onnx = self.profile_onnx_model(str(onnx_file))
        table_rows.append(self.generate_table_row(file.stem, t_onnx, t_engine, model_info))
        output.append(self.generate_results_dict(file.stem, t_onnx, t_engine, model_info))

    self.print_table(table_rows)
    return output

profile_onnx_model(onnx_file, eps=0.001)

ملفات تعريف و ONNX نموذج عن طريق تنفيذه عدة مرات وإرجاع المتوسط والانحراف المعياري للتشغيل تايمز.

شفرة المصدر في ultralytics/utils/benchmarks.py
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348349 350 351 352 353 354 355 356 357
def profile_onnx_model(self, onnx_file: str, eps: float = 1e-3):
    """Profiles an ONNX model by executing it multiple times and returns the mean and standard deviation of run
    times.
    """
    check_requirements("onnxruntime")
    import onnxruntime as ort

    # Session with either 'TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'
    sess_options = ort.SessionOptions()
    sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
    sess_options.intra_op_num_threads = 8  # Limit the number of threads
    sess = ort.InferenceSession(onnx_file, sess_options, providers=["CPUExecutionProvider"])

    input_tensor = sess.get_inputs()[0]
    input_type = input_tensor.type

    # Mapping ONNX datatype to numpy datatype
    if "float16" in input_type:
        input_dtype = np.float16
    elif "float" in input_type:
        input_dtype = np.float32
    elif "double" in input_type:
        input_dtype = np.float64
    elif "int64" in input_type:
        input_dtype = np.int64
    elif "int32" in input_type:
        input_dtype = np.int32
    else:
        raise ValueError(f"Unsupported ONNX datatype {input_type}")

    input_data = np.random.rand(*input_tensor.shape).astype(input_dtype)
    input_name = input_tensor.name
    output_name = sess.get_outputs()[0].name

    # Warmup runs
    elapsed = 0.0
    for _ in range(3):
        start_time = time.time()
        for _ in range(self.num_warmup_runs):
            sess.run([output_name], {input_name: input_data})
        elapsed = time.time() - start_time

    # Compute number of runs as higher of min_time or num_timed_runs
    num_runs = max(round(self.min_time / (elapsed + eps) * self.num_warmup_runs), self.num_timed_runs)

    # Timed runs
    run_times = []
    for _ in TQDM(range(num_runs), desc=onnx_file):
        start_time = time.time()
        sess.run([output_name], {input_name: input_data})
        run_times.append((time.time() - start_time) * 1000)  # Convert to milliseconds

    run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=5)  # sigma clipping
    return np.mean(run_times), np.std(run_times)

profile_tensorrt_model(engine_file, eps=0.001)

لمحة عن TensorRT نموذج ، يقيس متوسط وقت التشغيل والانحراف المعياري بين عمليات التشغيل.

شفرة المصدر في ultralytics/utils/benchmarks.py
275 276 277 278 279 280 281 282 283 284 285 286 287 288289 290 291 292 293 294 295 296297298 299300301302
def profile_tensorrt_model(self, engine_file: str, eps: float = 1e-3):
    """Profiles the TensorRT model, measuring average run time and standard deviation among runs."""
    if not self.trt or not Path(engine_file).is_file():
        return 0.0, 0.0

    # Model and input
    model = YOLO(engine_file)
    input_data = np.random.rand(self.imgsz, self.imgsz, 3).astype(np.float32)  # must be FP32

    # Warmup runs
    elapsed = 0.0
    for _ in range(3):
        start_time = time.time()
        for _ in range(self.num_warmup_runs):
            model(input_data, imgsz=self.imgsz, verbose=False)
        elapsed = time.time() - start_time

    # Compute number of runs as higher of min_time or num_timed_runs
    num_runs = max(round(self.min_time / (elapsed + eps) * self.num_warmup_runs), self.num_timed_runs * 50)

    # Timed runs
    run_times = []
    for _ in TQDM(range(num_runs), desc=engine_file):
        results = model(input_data, imgsz=self.imgsz, verbose=False)
        run_times.append(results[0].speed["inference"])  # Convert to milliseconds

    run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=3)  # sigma clipping
    return np.mean(run_times), np.std(run_times)



ultralytics.utils.benchmarks.benchmark(model=WEIGHTS_DIR / 'yolov8n.pt', data=None, imgsz=160, half=False, int8=False, device='cpu', verbose=False)

المعيار (أ) YOLO نموذج عبر تنسيقات مختلفة للسرعة والدقة.

البارامترات:

اسم نوع وصف افتراضي
model str | Path | optional

المسار إلى ملف النموذج أو الدليل. الافتراضي هو المسار (الإعدادات ['weights_dir']) / 'yolov8n.pt'.

WEIGHTS_DIR / 'yolov8n.pt'
data str

مجموعة البيانات للتقييم ، موروثة من TASK2DATA إذا لم يتم تمريرها. الافتراضي هو بلا.

None
imgsz int

حجم الصورة للمعيار. الافتراضي هو 160.

160
half bool

استخدم نصف الدقة للنموذج إذا كان صحيحا. الافتراضي هو خطأ.

False
int8 bool

استخدم int8-precision للنموذج إذا كان صحيحا. الافتراضي هو خطأ.

False
device str

جهاز لتشغيل المعيار ، إما "وحدة المعالجة المركزية" أو "cuda". الافتراضي هو "وحدة المعالجة المركزية".

'cpu'
verbose bool | float | optional

إذا كان صحيحا أو عائما ، فقم بتأكيد اجتياز المعايير بمقياس معين. الافتراضي هو خطأ.

False

ارجاع:

اسم نوع وصف
df DataFrame

إطار بيانات الباندا مع نتائج معيارية لكل تنسيق ، بما في ذلك حجم الملف ، متري ، ووقت الاستدلال.

مثل
from ultralytics.utils.benchmarks import benchmark

benchmark(model='yolov8n.pt', imgsz=640)
شفرة المصدر في ultralytics/utils/benchmarks.py
44 45 46 47 48  49 50 51 52 53 54 55 56 57 58 59 60  61 62 63  64 65 66 67     68       69 70  71  72 73  74 75 76  77         78  79 80 81 82 83 84 85  86 87 88  89 90 91 92 93 94 95 96 97 98  99  100    101 102 103  104   105 106  107  108 109    110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130131132 133 134 135 136 137 138139 140 141 142 143 144 145
def benchmark(
    model=WEIGHTS_DIR / "yolov8n.pt", data=None, imgsz=160, half=False, int8=False, device="cpu", verbose=False
):
    """
    Benchmark a YOLO model across different formats for speed and accuracy.

    Args:
        model (str | Path | optional): Path to the model file or directory. Default is
            Path(SETTINGS['weights_dir']) / 'yolov8n.pt'.
        data (str, optional): Dataset to evaluate on, inherited from TASK2DATA if not passed. Default is None.
        imgsz (int, optional): Image size for the benchmark. Default is 160.
        half (bool, optional): Use half-precision for the model if True. Default is False.
        int8 (bool, optional): Use int8-precision for the model if True. Default is False.
        device (str, optional): Device to run the benchmark on, either 'cpu' or 'cuda'. Default is 'cpu'.
        verbose (bool | float | optional): If True or a float, assert benchmarks pass with given metric.
            Default is False.

    Returns:
        df (pandas.DataFrame): A pandas DataFrame with benchmark results for each format, including file size,
            metric, and inference time.

    Example:
        ```python
        from ultralytics.utils.benchmarks import benchmark

        benchmark(model='yolov8n.pt', imgsz=640)
        ```
    """

    import pandas as pd

    pd.options.display.max_columns = 10
    pd.options.display.width = 120
    device = select_device(device, verbose=False)
    if isinstance(model, (str, Path)):
        model = YOLO(model)

    y = []
    t0 = time.time()
    for i, (name, format, suffix, cpu, gpu) in export_formats().iterrows():  # index, (name, format, suffix, CPU, GPU)
        emoji, filename = "❌", None  # export defaults
        try:
            # Checks
            if i == 9:
                assert LINUX, "Edge TPU export only supported on Linux"
            elif i == 7:
                assert model.task != "obb", "TensorFlow GraphDef not supported for OBB task"
            elif i in {5, 10}:  # CoreML and TF.js
                assert MACOS or LINUX, "export only supported on macOS and Linux"
            if i in {3, 5}:  # CoreML and OpenVINO
                assert not IS_PYTHON_3_12, "CoreML and OpenVINO not supported on Python 3.12"
            if "cpu" in device.type:
                assert cpu, "inference not supported on CPU"
            if "cuda" in device.type:
                assert gpu, "inference not supported on GPU"

            # Export
            if format == "-":
                filename = model.ckpt_path or model.cfg
                exported_model = model  # PyTorch format
            else:
                filename = model.export(imgsz=imgsz, format=format, half=half, int8=int8, device=device, verbose=False)
                exported_model = YOLO(filename, task=model.task)
                assert suffix in str(filename), "export failed"
            emoji = "❎"  # indicates export succeeded

            # Predict
            assert model.task != "pose" or i != 7, "GraphDef Pose inference is not supported"
            assert i not in (9, 10), "inference not supported"  # Edge TPU and TF.js are unsupported
            assert i != 5 or platform.system() == "Darwin", "inference only supported on macOS>=10.13"  # CoreML
            exported_model.predict(ASSETS / "bus.jpg", imgsz=imgsz, device=device, half=half)

            # Validate
            data = data or TASK2DATA[model.task]  # task to dataset, i.e. coco8.yaml for task=detect
            key = TASK2METRIC[model.task]  # task to metric, i.e. metrics/mAP50-95(B) for task=detect
            results = exported_model.val(
                data=data, batch=1, imgsz=imgsz, plots=False, device=device, half=half, int8=int8, verbose=False
            )
            metric, speed = results.results_dict[key], results.speed["inference"]
            y.append([name, "✅", round(file_size(filename), 1), round(metric, 4), round(speed, 2)])
        except Exception as e:
            if verbose:
                assert type(e) is AssertionError, f"Benchmark failure for {name}: {e}"
            LOGGER.warning(f"ERROR ❌️ Benchmark failure for {name}: {e}")
            y.append([name, emoji, round(file_size(filename), 1), None, None])  # mAP, t_inference

    # Print results
    check_yolo(device=device)  # print system info
    df = pd.DataFrame(y, columns=["Format", "Status❔", "Size (MB)", key, "Inference time (ms/im)"])

    name = Path(model.ckpt_path).name
    s = f"\nBenchmarks complete for {name} on {data} at imgsz={imgsz} ({time.time() - t0:.2f}s)\n{df}\n"
    LOGGER.info(s)
    with open("benchmarks.log", "a", errors="ignore", encoding="utf-8") as f:
        f.write(s)

    if verbose and isinstance(verbose, float):
        metrics = df[key].array  # values to compare to floor
        floor = verbose  # minimum metric floor to pass, i.e. = 0.29 mAP for YOLOv5n
        assert all(x > floor for x in metrics if pd.notna(x)), f"Benchmark failure: metric(s) < floor {floor}"

    return df





تم إنشاء 2023-11-12, اخر تحديث 2023-11-25
المؤلفون: جلين جوشر (3)