YOLOv8-16bit/ultralytics/yolo/engine/exporter.py

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Export a YOLOv5 PyTorch model to other formats. TensorFlow exports authored by https://github.com/zldrobit

Format                      | `export.py --include`         | Model
---                         | ---                           | ---
PyTorch                     | -                             | yolov8n.pt
TorchScript                 | `torchscript`                 | yolov8n.torchscript
ONNX                        | `onnx`                        | yolov8n.onnx
OpenVINO                    | `openvino`                    | yolov5s_openvino_model/
TensorRT                    | `engine`                      | yolov8n.engine
CoreML                      | `coreml`                      | yolov8n.mlmodel
TensorFlow SavedModel       | `saved_model`                 | yolov5s_saved_model/
TensorFlow GraphDef         | `pb`                          | yolov8n.pb
TensorFlow Lite             | `tflite`                      | yolov8n.tflite
TensorFlow Edge TPU         | `edgetpu`                     | yolov5s_edgetpu.tflite
TensorFlow.js               | `tfjs`                        | yolov5s_web_model/
PaddlePaddle                | `paddle`                      | yolov5s_paddle_model/

Requirements:
    $ pip install -r requirements.txt coremltools onnx onnx-simplifier onnxruntime openvino-dev tensorflow-cpu  # CPU
    $ pip install -r requirements.txt coremltools onnx onnx-simplifier onnxruntime-gpu openvino-dev tensorflow  # GPU

Usage:
    $ python export.py --weights yolov8n.pt --include torchscript onnx openvino engine coreml tflite ...

Inference:
    $ python detect.py --weights yolov8n.pt                 # PyTorch
                                 yolov8n.torchscript        # TorchScript
                                 yolov8n.onnx               # ONNX Runtime or OpenCV DNN with --dnn
                                 yolov5s_openvino_model     # OpenVINO
                                 yolov8n.engine             # TensorRT
                                 yolov8n.mlmodel            # CoreML (macOS-only)
                                 yolov5s_saved_model        # TensorFlow SavedModel
                                 yolov8n.pb                 # TensorFlow GraphDef
                                 yolov8n.tflite             # TensorFlow Lite
                                 yolov5s_edgetpu.tflite     # TensorFlow Edge TPU
                                 yolov5s_paddle_model       # PaddlePaddle

TensorFlow.js:
    $ cd .. && git clone https://github.com/zldrobit/tfjs-yolov5-example.git && cd tfjs-yolov5-example
    $ npm install
    $ ln -s ../../yolov5/yolov5s_web_model public/yolov5s_web_model
    $ npm start


from ultralytics import YOLO
model = YOLO().new('yolov8n.yaml')
results = model.export(format='onnx')
"""
import contextlib
import json
import os
import platform
import re
import subprocess
import time
import warnings
from copy import deepcopy
from pathlib import Path

import pandas as pd
import torch
from torch.utils.mobile_optimizer import optimize_for_mobile

from ultralytics.nn.modules import Detect, Segment
from ultralytics.nn.tasks import ClassificationModel, DetectionModel, SegmentationModel
from ultralytics.yolo.utils import LOGGER, ROOT, colorstr, get_default_args
from ultralytics.yolo.utils.checks import check_imgsz, check_requirements, check_version
from ultralytics.yolo.utils.files import file_size, yaml_save
from ultralytics.yolo.utils.ops import Profile
from ultralytics.yolo.utils.torch_utils import select_device, smart_inference_mode

MACOS = platform.system() == 'Darwin'  # macOS environment


def export_formats():
    # YOLOv5 export formats
    x = [
        ['PyTorch', '-', '.pt', True, True],
        ['TorchScript', 'torchscript', '.torchscript', True, True],
        ['ONNX', 'onnx', '.onnx', True, True],
        ['OpenVINO', 'openvino', '_openvino_model', True, False],
        ['TensorRT', 'engine', '.engine', False, True],
        ['CoreML', 'coreml', '.mlmodel', True, False],
        ['TensorFlow SavedModel', 'saved_model', '_saved_model', True, True],
        ['TensorFlow GraphDef', 'pb', '.pb', True, True],
        ['TensorFlow Lite', 'tflite', '.tflite', True, False],
        ['TensorFlow Edge TPU', 'edgetpu', '_edgetpu.tflite', False, False],
        ['TensorFlow.js', 'tfjs', '_web_model', False, False],
        ['PaddlePaddle', 'paddle', '_paddle_model', True, True],]
    return pd.DataFrame(x, columns=['Format', 'Argument', 'Suffix', 'CPU', 'GPU'])


def try_export(inner_func):
    # YOLOv5 export decorator, i..e @try_export
    inner_args = get_default_args(inner_func)

    def outer_func(*args, **kwargs):
        prefix = inner_args['prefix']
        try:
            with Profile() as dt:
                f, model = inner_func(*args, **kwargs)
            LOGGER.info(f'{prefix} export success ✅ {dt.t:.1f}s, saved as {f} ({file_size(f):.1f} MB)')
            return f, model
        except Exception as e:
            LOGGER.info(f'{prefix} export failure ❌ {dt.t:.1f}s: {e}')
            return None, None

    return outer_func


@try_export
def export_torchscript(model, im, file, optimize, prefix=colorstr('TorchScript:')):
    # YOLOv5 TorchScript model export
    LOGGER.info(f'\n{prefix} starting export with torch {torch.__version__}...')
    f = file.with_suffix('.torchscript')

    ts = torch.jit.trace(model, im, strict=False)
    d = {"shape": im.shape, "stride": int(max(model.stride)), "names": model.names}
    extra_files = {'config.txt': json.dumps(d)}  # torch._C.ExtraFilesMap()
    if optimize:  # https://pytorch.org/tutorials/recipes/mobile_interpreter.html
        optimize_for_mobile(ts)._save_for_lite_interpreter(str(f), _extra_files=extra_files)
    else:
        ts.save(str(f), _extra_files=extra_files)
    return f, None


@try_export
def export_onnx(model, im, file, opset, dynamic, simplify, prefix=colorstr('ONNX:')):
    # YOLOv5 ONNX export
    check_requirements('onnx>=1.12.0')
    import onnx  # noqa

    LOGGER.info(f'\n{prefix} starting export with onnx {onnx.__version__}...')
    f = file.with_suffix('.onnx')

    output_names = ['output0', 'output1'] if isinstance(model, SegmentationModel) else ['output0']
    if dynamic:
        dynamic = {'images': {0: 'batch', 2: 'height', 3: 'width'}}  # shape(1,3,640,640)
        if isinstance(model, SegmentationModel):
            dynamic['output0'] = {0: 'batch', 1: 'anchors'}  # shape(1,25200,85)
            dynamic['output1'] = {0: 'batch', 2: 'mask_height', 3: 'mask_width'}  # shape(1,32,160,160)
        elif isinstance(model, DetectionModel):
            dynamic['output0'] = {0: 'batch', 1: 'anchors'}  # shape(1,25200,85)

    torch.onnx.export(
        model.cpu() if dynamic else model,  # --dynamic only compatible with cpu
        im.cpu() if dynamic else im,
        f,
        verbose=False,
        opset_version=opset,
        do_constant_folding=True,  # WARNING: DNN inference with torch>=1.12 may require do_constant_folding=False
        input_names=['images'],
        output_names=output_names,
        dynamic_axes=dynamic or None)

    # Checks
    model_onnx = onnx.load(f)  # load onnx model
    onnx.checker.check_model(model_onnx)  # check onnx model

    # Metadata
    d = {'stride': int(max(model.stride)), 'names': model.names}
    for k, v in d.items():
        meta = model_onnx.metadata_props.add()
        meta.key, meta.value = k, str(v)
    onnx.save(model_onnx, f)

    # Simplify
    if simplify:
        try:
            cuda = torch.cuda.is_available()
            check_requirements(('onnxruntime-gpu' if cuda else 'onnxruntime', 'onnx-simplifier>=0.4.1'))
            import onnxsim

            LOGGER.info(f'{prefix} simplifying with onnx-simplifier {onnxsim.__version__}...')
            model_onnx, check = onnxsim.simplify(model_onnx)
            assert check, 'assert check failed'
            onnx.save(model_onnx, f)
        except Exception as e:
            LOGGER.info(f'{prefix} simplifier failure: {e}')
    return f, model_onnx


@try_export
def export_openvino(file, metadata, half, prefix=colorstr('OpenVINO:')):
    # YOLOv5 OpenVINO export
    check_requirements('openvino-dev')  # requires openvino-dev: https://pypi.org/project/openvino-dev/
    import openvino.inference_engine as ie  # noqa

    LOGGER.info(f'\n{prefix} starting export with openvino {ie.__version__}...')
    f = str(file).replace('.pt', f'_openvino_model{os.sep}')

    cmd = f"mo --input_model {file.with_suffix('.onnx')} --output_dir {f} --data_type {'FP16' if half else 'FP32'}"
    subprocess.run(cmd.split(), check=True, env=os.environ)  # export
    yaml_save(Path(f) / file.with_suffix('.yaml').name, metadata)  # add metadata.yaml
    return f, None


@try_export
def export_paddle(model, im, file, metadata, prefix=colorstr('PaddlePaddle:')):
    # YOLOv5 Paddle export
    check_requirements(('paddlepaddle', 'x2paddle'))
    import x2paddle  # noqa
    from x2paddle.convert import pytorch2paddle  # noqa

    LOGGER.info(f'\n{prefix} starting export with X2Paddle {x2paddle.__version__}...')
    f = str(file).replace('.pt', f'_paddle_model{os.sep}')

    pytorch2paddle(module=model, save_dir=f, jit_type='trace', input_examples=[im])  # export
    yaml_save(Path(f) / file.with_suffix('.yaml').name, metadata)  # add metadata.yaml
    return f, None


@try_export
def export_coreml(model, im, file, int8, half, prefix=colorstr('CoreML:')):
    # YOLOv5 CoreML export
    check_requirements('coremltools')
    import coremltools as ct  # noqa

    LOGGER.info(f'\n{prefix} starting export with coremltools {ct.__version__}...')
    f = file.with_suffix('.mlmodel')

    ts = torch.jit.trace(model, im, strict=False)  # TorchScript model
    ct_model = ct.convert(ts, inputs=[ct.ImageType('image', shape=im.shape, scale=1 / 255, bias=[0, 0, 0])])
    bits, mode = (8, 'kmeans_lut') if int8 else (16, 'linear') if half else (32, None)
    if bits < 32:
        if MACOS:  # quantization only supported on macOS
            ct_model = ct.models.neural_network.quantization_utils.quantize_weights(ct_model, bits, mode)
        else:
            LOGGER.info(f'{prefix} quantization only supported on macOS, skipping...')
    ct_model.save(f)
    return f, ct_model


@try_export
def export_engine(model, im, file, half, dynamic, simplify, workspace=4, verbose=False, prefix=colorstr('TensorRT:')):
    # YOLOv5 TensorRT export https://developer.nvidia.com/tensorrt
    assert im.device.type != 'cpu', 'export running on CPU but must be on GPU, i.e. `python export.py --device 0`'
    try:
        import tensorrt as trt
    except Exception:
        if platform.system() == 'Linux':
            check_requirements('nvidia-tensorrt', cmds='-U --index-url https://pypi.ngc.nvidia.com')
        import tensorrt as trt

    if trt.__version__[0] == '7':  # TensorRT 7 handling https://github.com/ultralytics/yolov5/issues/6012
        grid = model.model[-1].anchor_grid
        model.model[-1].anchor_grid = [a[..., :1, :1, :] for a in grid]
        export_onnx(model, im, file, 12, dynamic, simplify)  # opset 12
        model.model[-1].anchor_grid = grid
    else:  # TensorRT >= 8
        check_version(trt.__version__, '8.0.0', hard=True)  # require tensorrt>=8.0.0
        export_onnx(model, im, file, 12, dynamic, simplify)  # opset 12
    onnx = file.with_suffix('.onnx')

    LOGGER.info(f'\n{prefix} starting export with TensorRT {trt.__version__}...')
    assert onnx.exists(), f'failed to export ONNX file: {onnx}'
    f = file.with_suffix('.engine')  # TensorRT engine file
    logger = trt.Logger(trt.Logger.INFO)
    if verbose:
        logger.min_severity = trt.Logger.Severity.VERBOSE

    builder = trt.Builder(logger)
    config = builder.create_builder_config()
    config.max_workspace_size = workspace * 1 << 30
    # config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, workspace << 30)  # fix TRT 8.4 deprecation notice

    flag = (1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH))
    network = builder.create_network(flag)
    parser = trt.OnnxParser(network, logger)
    if not parser.parse_from_file(str(onnx)):
        raise RuntimeError(f'failed to load ONNX file: {onnx}')

    inputs = [network.get_input(i) for i in range(network.num_inputs)]
    outputs = [network.get_output(i) for i in range(network.num_outputs)]
    for inp in inputs:
        LOGGER.info(f'{prefix} input "{inp.name}" with shape{inp.shape} {inp.dtype}')
    for out in outputs:
        LOGGER.info(f'{prefix} output "{out.name}" with shape{out.shape} {out.dtype}')

    if dynamic:
        if im.shape[0] <= 1:
            LOGGER.warning(f"{prefix} WARNING ⚠️ --dynamic model requires maximum --batch-size argument")
        profile = builder.create_optimization_profile()
        for inp in inputs:
            profile.set_shape(inp.name, (1, *im.shape[1:]), (max(1, im.shape[0] // 2), *im.shape[1:]), im.shape)
        config.add_optimization_profile(profile)

    LOGGER.info(f'{prefix} building FP{16 if builder.platform_has_fast_fp16 and half else 32} engine as {f}')
    if builder.platform_has_fast_fp16 and half:
        config.set_flag(trt.BuilderFlag.FP16)
    with builder.build_engine(network, config) as engine, open(f, 'wb') as t:
        t.write(engine.serialize())
    return f, None


@try_export
def export_saved_model(model,
                       im,
                       file,
                       dynamic,
                       tf_nms=False,
                       agnostic_nms=False,
                       topk_per_class=100,
                       topk_all=100,
                       iou_thres=0.45,
                       conf_thres=0.25,
                       keras=False,
                       prefix=colorstr('TensorFlow SavedModel:')):
    # YOLOv5 TensorFlow SavedModel export
    try:
        import tensorflow as tf
    except Exception:
        check_requirements(f"tensorflow{'' if torch.cuda.is_available() else '-macos' if MACOS else '-cpu'}")
        import tensorflow as tf
    from models.tf import TFModel
    from tensorflow.python.framework.convert_to_constants import convert_variables_to_constants_v2  # noqa

    LOGGER.info(f'\n{prefix} starting export with tensorflow {tf.__version__}...')
    f = str(file).replace('.pt', '_saved_model')
    batch_size, ch, *imgsz = list(im.shape)  # BCHW

    tf_model = TFModel(cfg=model.yaml, model=model, nc=model.nc, imgsz=imgsz)
    im = tf.zeros((batch_size, *imgsz, ch))  # BHWC order for TensorFlow
    _ = tf_model.predict(im, tf_nms, agnostic_nms, topk_per_class, topk_all, iou_thres, conf_thres)
    inputs = tf.keras.Input(shape=(*imgsz, ch), batch_size=None if dynamic else batch_size)
    outputs = tf_model.predict(inputs, tf_nms, agnostic_nms, topk_per_class, topk_all, iou_thres, conf_thres)
    keras_model = tf.keras.Model(inputs=inputs, outputs=outputs)
    keras_model.trainable = False
    keras_model.summary()
    if keras:
        keras_model.save(f, save_format='tf')
    else:
        spec = tf.TensorSpec(keras_model.inputs[0].shape, keras_model.inputs[0].dtype)
        m = tf.function(lambda x: keras_model(x))  # full model
        m = m.get_concrete_function(spec)
        frozen_func = convert_variables_to_constants_v2(m)
        tfm = tf.Module()
        tfm.__call__ = tf.function(lambda x: frozen_func(x)[:4] if tf_nms else frozen_func(x), [spec])
        tfm.__call__(im)
        tf.saved_model.save(tfm,
                            f,
                            options=tf.saved_model.SaveOptions(experimental_custom_gradients=False) if check_version(
                                tf.__version__, '2.6') else tf.saved_model.SaveOptions())
    return f, keras_model


@try_export
def export_pb(keras_model, file, prefix=colorstr('TensorFlow GraphDef:')):
    # YOLOv5 TensorFlow GraphDef *.pb export https://github.com/leimao/Frozen_Graph_TensorFlow
    import tensorflow as tf  # noqa
    from tensorflow.python.framework.convert_to_constants import convert_variables_to_constants_v2  # noqa

    LOGGER.info(f'\n{prefix} starting export with tensorflow {tf.__version__}...')
    f = file.with_suffix('.pb')

    m = tf.function(lambda x: keras_model(x))  # full model
    m = m.get_concrete_function(tf.TensorSpec(keras_model.inputs[0].shape, keras_model.inputs[0].dtype))
    frozen_func = convert_variables_to_constants_v2(m)
    frozen_func.graph.as_graph_def()
    tf.io.write_graph(graph_or_graph_def=frozen_func.graph, logdir=str(f.parent), name=f.name, as_text=False)
    return f, None


@try_export
def export_tflite(keras_model, im, file, int8, data, nms, agnostic_nms, prefix=colorstr('TensorFlow Lite:')):
    # YOLOv5 TensorFlow Lite export
    import tensorflow as tf  # noqa

    LOGGER.info(f'\n{prefix} starting export with tensorflow {tf.__version__}...')
    batch_size, ch, *imgsz = list(im.shape)  # BCHW
    f = str(file).replace('.pt', '-fp16.tflite')

    converter = tf.lite.TFLiteConverter.from_keras_model(keras_model)
    converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS]
    converter.target_spec.supported_types = [tf.float16]
    converter.optimizations = [tf.lite.Optimize.DEFAULT]
    if int8:
        # from models.tf import representative_dataset_gen
        # dataset = LoadImages(check_dataset(check_yaml(data))['train'], imgsz=imgsz, auto=False)
        # converter.representative_dataset = lambda: representative_dataset_gen(dataset, ncalib=100)
        converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
        converter.target_spec.supported_types = []
        converter.inference_input_type = tf.uint8  # or tf.int8
        converter.inference_output_type = tf.uint8  # or tf.int8
        converter.experimental_new_quantizer = True
        f = str(file).replace('.pt', '-int8.tflite')
    if nms or agnostic_nms:
        converter.target_spec.supported_ops.append(tf.lite.OpsSet.SELECT_TF_OPS)

    tflite_model = converter.convert()
    open(f, "wb").write(tflite_model)
    return f, None


@try_export
def export_edgetpu(file, prefix=colorstr('Edge TPU:')):
    # YOLOv5 Edge TPU export https://coral.ai/docs/edgetpu/models-intro/
    cmd = 'edgetpu_compiler --version'
    help_url = 'https://coral.ai/docs/edgetpu/compiler/'
    assert platform.system() == 'Linux', f'export only supported on Linux. See {help_url}'
    if subprocess.run(f'{cmd} >/dev/null', shell=True).returncode != 0:
        LOGGER.info(f'\n{prefix} export requires Edge TPU compiler. Attempting install from {help_url}')
        sudo = subprocess.run('sudo --version >/dev/null', shell=True).returncode == 0  # sudo installed on system
        for c in (
                'curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -',
                'echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list',
                'sudo apt-get update', 'sudo apt-get install edgetpu-compiler'):
            subprocess.run(c if sudo else c.replace('sudo ', ''), shell=True, check=True)
    ver = subprocess.run(cmd, shell=True, capture_output=True, check=True).stdout.decode().split()[-1]

    LOGGER.info(f'\n{prefix} starting export with Edge TPU compiler {ver}...')
    f = str(file).replace('.pt', '-int8_edgetpu.tflite')  # Edge TPU model
    f_tfl = str(file).replace('.pt', '-int8.tflite')  # TFLite model

    cmd = f"edgetpu_compiler -s -d -k 10 --out_dir {file.parent} {f_tfl}"
    subprocess.run(cmd.split(), check=True)
    return f, None


@try_export
def export_tfjs(file, prefix=colorstr('TensorFlow.js:')):
    # YOLOv5 TensorFlow.js export
    check_requirements('tensorflowjs')
    import tensorflowjs as tfjs  # noqa

    LOGGER.info(f'\n{prefix} starting export with tensorflowjs {tfjs.__version__}...')
    f = str(file).replace('.pt', '_web_model')  # js dir
    f_pb = file.with_suffix('.pb')  # *.pb path
    f_json = f'{f}/model.json'  # *.json path

    cmd = f'tensorflowjs_converter --input_format=tf_frozen_model ' \
          f'--output_node_names=Identity,Identity_1,Identity_2,Identity_3 {f_pb} {f}'
    subprocess.run(cmd.split())

    json = Path(f_json).read_text()
    with open(f_json, 'w') as j:  # sort JSON Identity_* in ascending order
        subst = re.sub(
            r'{"outputs": {"Identity.?.?": {"name": "Identity.?.?"}, '
            r'"Identity.?.?": {"name": "Identity.?.?"}, '
            r'"Identity.?.?": {"name": "Identity.?.?"}, '
            r'"Identity.?.?": {"name": "Identity.?.?"}}}', r'{"outputs": {"Identity": {"name": "Identity"}, '
            r'"Identity_1": {"name": "Identity_1"}, '
            r'"Identity_2": {"name": "Identity_2"}, '
            r'"Identity_3": {"name": "Identity_3"}}}', json)
        j.write(subst)
    return f, None


def add_tflite_metadata(file, metadata, num_outputs):
    # Add metadata to *.tflite models per https://www.tensorflow.org/lite/models/convert/metadata
    with contextlib.suppress(ImportError):
        # check_requirements('tflite_support')
        from tflite_support import flatbuffers  # noqa
        from tflite_support import metadata as _metadata  # noqa
        from tflite_support import metadata_schema_py_generated as _metadata_fb  # noqa

        tmp_file = Path('/tmp/meta.txt')
        with open(tmp_file, 'w') as meta_f:
            meta_f.write(str(metadata))

        model_meta = _metadata_fb.ModelMetadataT()
        label_file = _metadata_fb.AssociatedFileT()
        label_file.name = tmp_file.name
        model_meta.associatedFiles = [label_file]

        subgraph = _metadata_fb.SubGraphMetadataT()
        subgraph.inputTensorMetadata = [_metadata_fb.TensorMetadataT()]
        subgraph.outputTensorMetadata = [_metadata_fb.TensorMetadataT()] * num_outputs
        model_meta.subgraphMetadata = [subgraph]

        b = flatbuffers.Builder(0)
        b.Finish(model_meta.Pack(b), _metadata.MetadataPopulator.METADATA_FILE_IDENTIFIER)
        metadata_buf = b.Output()

        populator = _metadata.MetadataPopulator.with_model_file(file)
        populator.load_metadata_buffer(metadata_buf)
        populator.load_associated_files([str(tmp_file)])
        populator.populate()
        tmp_file.unlink()


@smart_inference_mode()
def export_model(
        model,  # model
        file=ROOT / 'yolov8n.pt',
        data=ROOT / 'data/coco128.yaml',  # 'dataset.yaml path'
        imgsz=(640, 640),  # image (height, width)
        batch_size=1,  # batch size
        device=torch.device('cpu'),  # cuda device, i.e. 0 or 0,1,2,3 or cpu
        format='onnx',  # export format
        half=False,  # FP16 half-precision export
        keras=False,  # use Keras
        optimize=False,  # TorchScript: optimize for mobile
        int8=False,  # CoreML/TF INT8 quantization
        dynamic=False,  # ONNX/TF/TensorRT: dynamic axes
        simplify=False,  # ONNX: simplify model
        opset=17,  # ONNX: opset version
        verbose=False,  # TensorRT: verbose log
        workspace=4,  # TensorRT: workspace size (GB)
        nms=False,  # TF: add NMS to model
        agnostic_nms=False,  # TF: add agnostic NMS to model
        topk_per_class=100,  # TF.js NMS: topk per class to keep
        topk_all=100,  # TF.js NMS: topk for all classes to keep
        iou_thres=0.45,  # TF.js NMS: IoU threshold
        conf_thres=0.25,  # TF.js NMS: confidence threshold
):
    t = time.time()
    format = format.lower()  # to lowercase
    fmts = tuple(export_formats()['Argument'][1:])  # available export formats
    flags = [x == format for x in fmts]
    assert sum(flags), f'ERROR: Invalid format={format}, valid formats are {fmts}'
    jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle = flags  # export booleans

    # Load PyTorch model
    device = select_device(device)
    if half:
        assert device.type != 'cpu' or coreml, '--half only compatible with GPU export, i.e. use --device 0'
        assert not dynamic, '--half not compatible with --dynamic, i.e. use either --half or --dynamic but not both'
    model = deepcopy(model).fuse()  # load FP32 model

    # Checks
    if isinstance(imgsz, int):
        imgsz = [imgsz]
    imgsz *= 2 if len(imgsz) == 1 else 1  # expand
    if optimize:
        assert device.type == 'cpu', '--optimize not compatible with cuda devices, i.e. use --device cpu'

    # Input
    gs = int(max(model.stride))  # grid size (max stride)
    imgsz = [check_imgsz(x, gs) for x in imgsz]  # verify img_size are gs-multiples
    im = torch.zeros(batch_size, 3, *imgsz).to(device)  # image size(1,3,320,192) BCHW iDetection

    # Update model
    model.eval()
    for k, m in model.named_modules():
        if isinstance(m, (Detect, Segment)):
            m.dynamic = dynamic
            m.export = True

    for _ in range(2):
        y = model(im)  # dry runs
    if half and not coreml:
        im, model = im.half(), model.half()  # to FP16
    shape = tuple((y[0] if isinstance(y, tuple) else y).shape)  # model output shape
    metadata = {'stride': int(max(model.stride)), 'names': model.names}  # model metadata
    LOGGER.info(f"\n{colorstr('PyTorch:')} starting from {file} with output shape {shape} ({file_size(file):.1f} MB)")

    # Warnings
    warnings.filterwarnings('ignore', category=torch.jit.TracerWarning)  # suppress TracerWarning
    warnings.filterwarnings('ignore', category=UserWarning)  # suppress shape prim::Constant type missing ONNX warning
    warnings.filterwarnings('ignore', category=DeprecationWarning)  # suppress CoreML np.bool deprecation warning

    # Exports
    f = [''] * len(fmts)  # exported filenames
    if jit:  # TorchScript
        f[0], _ = export_torchscript(model, im, file, optimize)
    if engine:  # TensorRT required before ONNX
        f[1], _ = export_engine(model, im, file, half, dynamic, simplify, workspace, verbose)
    if onnx or xml:  # OpenVINO requires ONNX
        f[2], _ = export_onnx(model, im, file, opset, dynamic, simplify)
    if xml:  # OpenVINO
        f[3], _ = export_openvino(file, metadata, half)
    if coreml:  # CoreML
        f[4], _ = export_coreml(model, im, file, int8, half)
    if any((saved_model, pb, tflite, edgetpu, tfjs)):  # TensorFlow formats
        assert not tflite or not tfjs, 'TFLite and TF.js models must be exported separately, please pass only one type.'
        assert not isinstance(model, ClassificationModel), 'ClassificationModel export to TF formats not yet supported.'
        f[5], s_model = export_saved_model(model.cpu(),
                                           im,
                                           file,
                                           dynamic,
                                           tf_nms=nms or agnostic_nms or tfjs,
                                           agnostic_nms=agnostic_nms or tfjs,
                                           topk_per_class=topk_per_class,
                                           topk_all=topk_all,
                                           iou_thres=iou_thres,
                                           conf_thres=conf_thres,
                                           keras=keras)
        if pb or tfjs:  # pb prerequisite to tfjs
            f[6], _ = export_pb(s_model, file)
        if tflite or edgetpu:
            f[7], _ = export_tflite(s_model, im, file, int8 or edgetpu, data=data, nms=nms, agnostic_nms=agnostic_nms)
            if edgetpu:
                f[8], _ = export_edgetpu(file)
            add_tflite_metadata(f[8] or f[7], metadata, num_outputs=len(s_model.outputs))
        if tfjs:
            f[9], _ = export_tfjs(file)
    if paddle:  # PaddlePaddle
        f[10], _ = export_paddle(model, im, file, metadata)

    # Finish
    f = [str(x) for x in f if x]  # filter out '' and None
    if any(f):
        cls, det, seg = (isinstance(model, x) for x in (ClassificationModel, DetectionModel, SegmentationModel))  # type
        det &= not seg  # segmentation models inherit from SegmentationModel(DetectionModel)
        dir = Path('segment' if seg else 'classify' if cls else '')
        h = '--half' if half else ''  # --half FP16 inference arg
        s = "# WARNING ⚠️ ClassificationModel not yet supported for PyTorch Hub AutoShape inference" if cls else \
            "# WARNING ⚠️ SegmentationModel not yet supported for PyTorch Hub AutoShape inference" if seg else ''
        LOGGER.info(f'\nExport complete ({time.time() - t:.1f}s)'
                    f"\nResults saved to {colorstr('bold', file.parent.resolve())}"
                    f"\nDetect:          python {dir / 'predict.py'} --weights {f[-1]} {h}"
                    f"\nValidate:        python {dir / 'val.py'} --weights {f[-1]} {h}"
                    f"\nPyTorch Hub:     model = torch.hub.load('ultralytics/yolov5', 'custom', '{f[-1]}')  {s}"
                    f"\nVisualize:       https://netron.app")
    return f  # return list of exported files/dirs