Skip to content

Exporter

Exporter API Reference

Exporter

A class for exporting a model.

Attributes:

Name Type Description
args SimpleNamespace

Configuration for the exporter.

save_dir Path

Directory to save results.

Source code in ultralytics/yolo/engine/exporter.py
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
class Exporter:
    """
    Exporter

    A class for exporting a model.

    Attributes:
        args (SimpleNamespace): Configuration for the exporter.
        save_dir (Path): Directory to save results.
    """

    def __init__(self, cfg=DEFAULT_CFG, overrides=None):
        """
        Initializes the Exporter class.

        Args:
            cfg (str, optional): Path to a configuration file. Defaults to DEFAULT_CFG.
            overrides (dict, optional): Configuration overrides. Defaults to None.
        """
        self.args = get_cfg(cfg, overrides)
        self.callbacks = defaultdict(list, callbacks.default_callbacks)  # add callbacks
        callbacks.add_integration_callbacks(self)

    @smart_inference_mode()
    def __call__(self, model=None):
        self.run_callbacks("on_export_start")
        t = time.time()
        format = self.args.format.lower()  # to lowercase
        if format in {'tensorrt', 'trt'}:  # engine aliases
            format = 'engine'
        fmts = tuple(export_formats()['Argument'][1:])  # available export formats
        flags = [x == format for x in fmts]
        if sum(flags) != 1:
            raise ValueError(f"Invalid export format='{format}'. Valid formats are {fmts}")
        jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle = flags  # export booleans

        # Load PyTorch model
        self.device = select_device('cpu' if self.args.device is None else self.args.device)
        if self.args.half:
            if self.device.type == 'cpu' and not coreml and not xml:
                LOGGER.info('half=True only compatible with GPU or CoreML export, i.e. use device=0 or format=coreml')
                self.args.half = False
            assert not self.args.dynamic, '--half not compatible with --dynamic, i.e. use either --half or --dynamic'

        # Checks
        model.names = check_class_names(model.names)
        # if self.args.batch == model.args['batch_size']:  # user has not modified training batch_size
        self.args.batch = 1
        self.imgsz = check_imgsz(self.args.imgsz, stride=model.stride, min_dim=2)  # check image size
        if model.task == 'classify':
            self.args.nms = self.args.agnostic_nms = False
        if self.args.optimize:
            assert self.device.type == 'cpu', '--optimize not compatible with cuda devices, i.e. use --device cpu'

        # Input
        im = torch.zeros(self.args.batch, 3, *self.imgsz).to(self.device)
        file = Path(getattr(model, 'pt_path', None) or getattr(model, 'yaml_file', None) or model.yaml['yaml_file'])
        if file.suffix == '.yaml':
            file = Path(file.name)

        # Update model
        model = deepcopy(model).to(self.device)
        for p in model.parameters():
            p.requires_grad = False
        model.eval()
        model.float()
        model = model.fuse()
        for k, m in model.named_modules():
            if isinstance(m, (Detect, Segment)):
                m.dynamic = self.args.dynamic
                m.export = True

        y = None
        for _ in range(2):
            y = model(im)  # dry runs
        if self.args.half and not coreml and not xml:
            im, model = im.half(), model.half()  # to FP16

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

        # Assign
        self.im = im
        self.model = model
        self.file = file
        self.output_shape = tuple(y.shape) if isinstance(y, torch.Tensor) else tuple(tuple(x.shape) for x in y)
        self.pretty_name = self.file.stem.replace('yolo', 'YOLO')
        self.metadata = {
            'description': f"Ultralytics {self.pretty_name} model trained on {self.model.args['data']}",
            'author': 'Ultralytics',
            'license': 'GPL-3.0 https://ultralytics.com/license',
            'version': ultralytics.__version__,
            'stride': int(max(model.stride)),
            'names': model.names}  # model metadata

        LOGGER.info(f"\n{colorstr('PyTorch:')} starting from {file} with input shape {tuple(im.shape)} BCHW and "
                    f"output shape(s) {self.output_shape} ({file_size(file):.1f} MB)")

        # Exports
        f = [''] * len(fmts)  # exported filenames
        if jit:  # TorchScript
            f[0], _ = self._export_torchscript()
        if engine:  # TensorRT required before ONNX
            f[1], _ = self._export_engine()
        if onnx or xml:  # OpenVINO requires ONNX
            f[2], _ = self._export_onnx()
        if xml:  # OpenVINO
            f[3], _ = self._export_openvino()
        if coreml:  # CoreML
            f[4], _ = self._export_coreml()
        if any((saved_model, pb, tflite, edgetpu, tfjs)):  # TensorFlow formats
            LOGGER.warning('WARNING ⚠️ YOLOv8 TensorFlow export support is still under development. '
                           'Please consider contributing to the effort if you have TF expertise. Thank you!')
            nms = False
            f[5], s_model = self._export_saved_model(nms=nms or self.args.agnostic_nms or tfjs,
                                                     agnostic_nms=self.args.agnostic_nms or tfjs)

            debug = False
            if debug:
                if pb or tfjs:  # pb prerequisite to tfjs
                    f[6], _ = self._export_pb(s_model)
                if tflite or edgetpu:
                    f[7], _ = self._export_tflite(s_model,
                                                  int8=self.args.int8 or edgetpu,
                                                  data=self.args.data,
                                                  nms=nms,
                                                  agnostic_nms=self.args.agnostic_nms)
                    if edgetpu:
                        f[8], _ = self._export_edgetpu()
                    self._add_tflite_metadata(f[8] or f[7], num_outputs=len(self.output_shape))
                if tfjs:
                    f[9], _ = self._export_tfjs()
        if paddle:  # PaddlePaddle
            f[10], _ = self._export_paddle()

        # Finish
        f = [str(x) for x in f if x]  # filter out '' and None
        if any(f):
            s = "-WARNING ⚠️ not yet supported for YOLOv8 exported models"
            LOGGER.info(f'\nExport complete ({time.time() - t:.1f}s)'
                        f"\nResults saved to {colorstr('bold', file.parent.resolve())}"
                        f"\nPredict:         yolo task={model.task} mode=predict model={f[-1]} {s}"
                        f"\nValidate:        yolo task={model.task} mode=val model={f[-1]} {s}"
                        f"\nVisualize:       https://netron.app")

        self.run_callbacks("on_export_end")
        return f  # return list of exported files/dirs

    @try_export
    def _export_torchscript(self, prefix=colorstr('TorchScript:')):
        # YOLOv8 TorchScript model export
        LOGGER.info(f'\n{prefix} starting export with torch {torch.__version__}...')
        f = self.file.with_suffix('.torchscript')

        ts = torch.jit.trace(self.model, self.im, strict=False)
        d = {"shape": self.im.shape, "stride": int(max(self.model.stride)), "names": self.model.names}
        extra_files = {'config.txt': json.dumps(d)}  # torch._C.ExtraFilesMap()
        if self.args.optimize:  # https://pytorch.org/tutorials/recipes/mobile_interpreter.html
            LOGGER.info(f'{prefix} optimizing for mobile...')
            from torch.utils.mobile_optimizer import optimize_for_mobile
            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(self, prefix=colorstr('ONNX:')):
        # YOLOv8 ONNX export
        check_requirements('onnx>=1.12.0')
        import onnx  # noqa

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

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

        torch.onnx.export(
            self.model.cpu() if dynamic else self.model,  # --dynamic only compatible with cpu
            self.im.cpu() if dynamic else self.im,
            f,
            verbose=False,
            opset_version=self.args.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(self.model.stride)), 'names': self.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 self.args.simplify:
            try:
                check_requirements('onnxsim')
                import onnxsim

                LOGGER.info(f'{prefix} simplifying with onnx-simplifier {onnxsim.__version__}...')
                subprocess.run(f'onnxsim {f} {f}', shell=True)
            except Exception as e:
                LOGGER.info(f'{prefix} simplifier failure: {e}')
        return f, model_onnx

    @try_export
    def _export_openvino(self, prefix=colorstr('OpenVINO:')):
        # YOLOv8 OpenVINO export
        check_requirements('openvino-dev>=2022.3')  # requires openvino-dev: https://pypi.org/project/openvino-dev/
        import openvino.runtime as ov  # noqa
        from openvino.tools import mo  # noqa

        LOGGER.info(f'\n{prefix} starting export with openvino {ov.__version__}...')
        f = str(self.file).replace(self.file.suffix, f'_openvino_model{os.sep}')
        f_onnx = self.file.with_suffix('.onnx')
        f_ov = str(Path(f) / self.file.with_suffix('.xml').name)

        ov_model = mo.convert_model(f_onnx,
                                    model_name=self.pretty_name,
                                    framework="onnx",
                                    compress_to_fp16=self.args.half)  # export
        ov.serialize(ov_model, f_ov)  # save
        yaml_save(Path(f) / self.file.with_suffix('.yaml').name, self.metadata)  # add metadata.yaml
        return f, None

    @try_export
    def _export_paddle(self, prefix=colorstr('PaddlePaddle:')):
        # YOLOv8 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(self.file).replace(self.file.suffix, f'_paddle_model{os.sep}')

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

    @try_export
    def _export_coreml(self, prefix=colorstr('CoreML:')):
        # YOLOv8 CoreML export
        check_requirements('coremltools>=6.0')
        import coremltools as ct  # noqa

        class iOSModel(torch.nn.Module):
            # Wrap an Ultralytics YOLO model for iOS export
            def __init__(self, model, im):
                super().__init__()
                b, c, h, w = im.shape  # batch, channel, height, width
                self.model = model
                self.nc = len(model.names)  # number of classes
                if w == h:
                    self.normalize = 1.0 / w  # scalar
                else:
                    self.normalize = torch.tensor([1.0 / w, 1.0 / h, 1.0 / w, 1.0 / h])  # broadcast (slower, smaller)

            def forward(self, x):
                xywh, cls = self.model(x)[0].transpose(0, 1).split((4, self.nc), 1)
                return cls, xywh * self.normalize  # confidence (3780, 80), coordinates (3780, 4)

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

        if self.model.task == 'classify':
            bias = [-x for x in IMAGENET_MEAN]
            scale = 1 / 255 / (sum(IMAGENET_STD) / 3)
            classifier_config = ct.ClassifierConfig(list(self.model.names.values()))
        else:
            bias = [0.0, 0.0, 0.0]
            scale = 1 / 255
            classifier_config = None
        model = iOSModel(self.model, self.im).eval() if self.args.nms else self.model
        ts = torch.jit.trace(model, self.im, strict=False)  # TorchScript model
        ct_model = ct.convert(ts,
                              inputs=[ct.ImageType('image', shape=self.im.shape, scale=scale, bias=bias)],
                              classifier_config=classifier_config)
        bits, mode = (8, 'kmeans_lut') if self.args.int8 else (16, 'linear') if self.args.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...')
        if self.args.nms:
            ct_model = self._pipeline_coreml(ct_model)

        ct_model.short_description = self.metadata['description']
        ct_model.author = self.metadata['author']
        ct_model.license = self.metadata['license']
        ct_model.version = self.metadata['version']
        ct_model.save(str(f))
        return f, ct_model

    @try_export
    def _export_engine(self, workspace=4, verbose=False, prefix=colorstr('TensorRT:')):
        # YOLOv8 TensorRT export https://developer.nvidia.com/tensorrt
        assert self.im.device.type != 'cpu', "export running on CPU but must be on GPU, i.e. use 'device=0'"
        try:
            import tensorrt as trt  # noqa
        except ImportError:
            if platform.system() == 'Linux':
                check_requirements('nvidia-tensorrt', cmds='-U --index-url https://pypi.ngc.nvidia.com')
            import tensorrt as trt  # noqa

        check_version(trt.__version__, '7.0.0', hard=True)  # require tensorrt>=8.0.0
        self._export_onnx()
        onnx = self.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 = self.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 self.args.dynamic:
            shape = self.im.shape
            if 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, *shape[1:]), (max(1, shape[0] // 2), *shape[1:]), shape)
            config.add_optimization_profile(profile)

        LOGGER.info(
            f'{prefix} building FP{16 if builder.platform_has_fast_fp16 and self.args.half else 32} engine as {f}')
        if builder.platform_has_fast_fp16 and self.args.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(self,
                            nms=False,
                            agnostic_nms=False,
                            topk_per_class=100,
                            topk_all=100,
                            iou_thres=0.45,
                            conf_thres=0.25,
                            prefix=colorstr('TensorFlow SavedModel:')):

        # YOLOv8 TensorFlow SavedModel export
        try:
            import tensorflow as tf  # noqa
        except ImportError:
            check_requirements(f"tensorflow{'' if torch.cuda.is_available() else '-macos' if MACOS else '-cpu'}")
            import tensorflow as tf  # noqa
        check_requirements(("onnx", "onnx2tf", "sng4onnx", "onnxsim", "onnx_graphsurgeon"),
                           cmds="--extra-index-url https://pypi.ngc.nvidia.com ")

        LOGGER.info(f'\n{prefix} starting export with tensorflow {tf.__version__}...')
        f = str(self.file).replace(self.file.suffix, '_saved_model')

        # Export to ONNX
        self._export_onnx()
        onnx = self.file.with_suffix('.onnx')

        # Export to TF SavedModel
        subprocess.run(f'onnx2tf -i {onnx} --output_signaturedefs -o {f}', shell=True)

        # Load saved_model
        keras_model = tf.saved_model.load(f, tags=None, options=None)

        return f, keras_model

    @try_export
    def _export_saved_model_OLD(self,
                                nms=False,
                                agnostic_nms=False,
                                topk_per_class=100,
                                topk_all=100,
                                iou_thres=0.45,
                                conf_thres=0.25,
                                prefix=colorstr('TensorFlow SavedModel:')):
        # YOLOv8 TensorFlow SavedModel export
        try:
            import tensorflow as tf  # noqa
        except ImportError:
            check_requirements(f"tensorflow{'' if torch.cuda.is_available() else '-macos' if MACOS else '-cpu'}")
            import tensorflow as tf  # noqa
        # 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(self.file).replace(self.file.suffix, '_saved_model')
        batch_size, ch, *imgsz = list(self.im.shape)  # BCHW

        tf_models = None  # TODO: no TF modules available
        tf_model = tf_models.TFModel(cfg=self.model.yaml, model=self.model.cpu(), nc=self.model.nc, imgsz=imgsz)
        im = tf.zeros((batch_size, *imgsz, ch))  # BHWC order for TensorFlow
        _ = tf_model.predict(im, nms, agnostic_nms, topk_per_class, topk_all, iou_thres, conf_thres)
        inputs = tf.keras.Input(shape=(*imgsz, ch), batch_size=None if self.args.dynamic else batch_size)
        outputs = tf_model.predict(inputs, 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 self.args.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 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(self, keras_model, prefix=colorstr('TensorFlow GraphDef:')):
        # YOLOv8 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 = self.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(self, keras_model, int8, data, nms, agnostic_nms, prefix=colorstr('TensorFlow Lite:')):
        # YOLOv8 TensorFlow Lite export
        import tensorflow as tf  # noqa

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

            def representative_dataset_gen(dataset, n_images=100):
                # Dataset generator for use with converter.representative_dataset, returns a generator of np arrays
                for n, (path, img, im0s, vid_cap, string) in enumerate(dataset):
                    im = np.transpose(img, [1, 2, 0])
                    im = np.expand_dims(im, axis=0).astype(np.float32)
                    im /= 255
                    yield [im]
                    if n >= n_images:
                        break

            dataset = LoadImages(check_det_dataset(check_yaml(data))['train'], imgsz=imgsz, auto=False)
            converter.representative_dataset = lambda: representative_dataset_gen(dataset, n_images=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(self.file).replace(self.file.suffix, '-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(self, prefix=colorstr('Edge TPU:')):
        # YOLOv8 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" | '  # no comma
                    '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(self.file).replace(self.file.suffix, '-int8_edgetpu.tflite')  # Edge TPU model
        f_tfl = str(self.file).replace(self.file.suffix, '-int8.tflite')  # TFLite model

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

    @try_export
    def _export_tfjs(self, prefix=colorstr('TensorFlow.js:')):
        # YOLOv8 TensorFlow.js export
        check_requirements('tensorflowjs')
        import tensorflowjs as tfjs  # noqa

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

        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"}}}', f_json.read_text())
            j.write(subst)
        return f, None

    def _add_tflite_metadata(self, file, 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(self.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()

    def _pipeline_coreml(self, model, prefix=colorstr('CoreML Pipeline:')):
        # YOLOv8 CoreML pipeline
        import coremltools as ct  # noqa

        LOGGER.info(f'{prefix} starting pipeline with coremltools {ct.__version__}...')
        batch_size, ch, h, w = list(self.im.shape)  # BCHW

        # Output shapes
        spec = model.get_spec()
        out0, out1 = iter(spec.description.output)
        if MACOS:
            from PIL import Image
            img = Image.new('RGB', (w, h))  # img(192 width, 320 height)
            # img = torch.zeros((*opt.img_size, 3)).numpy()  # img size(320,192,3) iDetection
            out = model.predict({'image': img})
            out0_shape = out[out0.name].shape
            out1_shape = out[out1.name].shape
        else:  # linux and windows can not run model.predict(), get sizes from pytorch output y
            out0_shape = self.output_shape[2], self.output_shape[1] - 4  # (3780, 80)
            out1_shape = self.output_shape[2], 4  # (3780, 4)

        # Checks
        names = self.metadata['names']
        nx, ny = spec.description.input[0].type.imageType.width, spec.description.input[0].type.imageType.height
        na, nc = out0_shape
        # na, nc = out0.type.multiArrayType.shape  # number anchors, classes
        assert len(names) == nc, f'{len(names)} names found for nc={nc}'  # check

        # Define output shapes (missing)
        out0.type.multiArrayType.shape[:] = out0_shape  # (3780, 80)
        out1.type.multiArrayType.shape[:] = out1_shape  # (3780, 4)
        # spec.neuralNetwork.preprocessing[0].featureName = '0'

        # Flexible input shapes
        # from coremltools.models.neural_network import flexible_shape_utils
        # s = [] # shapes
        # s.append(flexible_shape_utils.NeuralNetworkImageSize(320, 192))
        # s.append(flexible_shape_utils.NeuralNetworkImageSize(640, 384))  # (height, width)
        # flexible_shape_utils.add_enumerated_image_sizes(spec, feature_name='image', sizes=s)
        # r = flexible_shape_utils.NeuralNetworkImageSizeRange()  # shape ranges
        # r.add_height_range((192, 640))
        # r.add_width_range((192, 640))
        # flexible_shape_utils.update_image_size_range(spec, feature_name='image', size_range=r)

        # Print
        # print(spec.description)

        # Model from spec
        model = ct.models.MLModel(spec)

        # 3. Create NMS protobuf
        nms_spec = ct.proto.Model_pb2.Model()
        nms_spec.specificationVersion = 5
        for i in range(2):
            decoder_output = model._spec.description.output[i].SerializeToString()
            nms_spec.description.input.add()
            nms_spec.description.input[i].ParseFromString(decoder_output)
            nms_spec.description.output.add()
            nms_spec.description.output[i].ParseFromString(decoder_output)

        nms_spec.description.output[0].name = 'confidence'
        nms_spec.description.output[1].name = 'coordinates'

        output_sizes = [nc, 4]
        for i in range(2):
            ma_type = nms_spec.description.output[i].type.multiArrayType
            ma_type.shapeRange.sizeRanges.add()
            ma_type.shapeRange.sizeRanges[0].lowerBound = 0
            ma_type.shapeRange.sizeRanges[0].upperBound = -1
            ma_type.shapeRange.sizeRanges.add()
            ma_type.shapeRange.sizeRanges[1].lowerBound = output_sizes[i]
            ma_type.shapeRange.sizeRanges[1].upperBound = output_sizes[i]
            del ma_type.shape[:]

        nms = nms_spec.nonMaximumSuppression
        nms.confidenceInputFeatureName = out0.name  # 1x507x80
        nms.coordinatesInputFeatureName = out1.name  # 1x507x4
        nms.confidenceOutputFeatureName = 'confidence'
        nms.coordinatesOutputFeatureName = 'coordinates'
        nms.iouThresholdInputFeatureName = 'iouThreshold'
        nms.confidenceThresholdInputFeatureName = 'confidenceThreshold'
        nms.iouThreshold = 0.45
        nms.confidenceThreshold = 0.25
        nms.pickTop.perClass = True
        nms.stringClassLabels.vector.extend(names.values())
        nms_model = ct.models.MLModel(nms_spec)

        # 4. Pipeline models together
        pipeline = ct.models.pipeline.Pipeline(input_features=[('image', ct.models.datatypes.Array(3, ny, nx)),
                                                               ('iouThreshold', ct.models.datatypes.Double()),
                                                               ('confidenceThreshold', ct.models.datatypes.Double())],
                                               output_features=['confidence', 'coordinates'])
        pipeline.add_model(model)
        pipeline.add_model(nms_model)

        # Correct datatypes
        pipeline.spec.description.input[0].ParseFromString(model._spec.description.input[0].SerializeToString())
        pipeline.spec.description.output[0].ParseFromString(nms_model._spec.description.output[0].SerializeToString())
        pipeline.spec.description.output[1].ParseFromString(nms_model._spec.description.output[1].SerializeToString())

        # Update metadata
        pipeline.spec.specificationVersion = 5
        pipeline.spec.description.metadata.userDefined.update({
            'IoU threshold': str(nms.iouThreshold),
            'Confidence threshold': str(nms.confidenceThreshold)})

        # Save the model
        model = ct.models.MLModel(pipeline.spec)
        model.input_description['image'] = 'Input image'
        model.input_description['iouThreshold'] = f'(optional) IOU threshold override (default: {nms.iouThreshold})'
        model.input_description['confidenceThreshold'] = \
            f'(optional) Confidence threshold override (default: {nms.confidenceThreshold})'
        model.output_description['confidence'] = 'Boxes × Class confidence (see user-defined metadata "classes")'
        model.output_description['coordinates'] = 'Boxes × [x, y, width, height] (relative to image size)'
        LOGGER.info(f'{prefix} pipeline success')
        return model

    def run_callbacks(self, event: str):
        for callback in self.callbacks.get(event, []):
            callback(self)

__init__(cfg=DEFAULT_CFG, overrides=None)

Initializes the Exporter class.

Parameters:

Name Type Description Default
cfg str

Path to a configuration file. Defaults to DEFAULT_CFG.

DEFAULT_CFG
overrides dict

Configuration overrides. Defaults to None.

None
Source code in ultralytics/yolo/engine/exporter.py
def __init__(self, cfg=DEFAULT_CFG, overrides=None):
    """
    Initializes the Exporter class.

    Args:
        cfg (str, optional): Path to a configuration file. Defaults to DEFAULT_CFG.
        overrides (dict, optional): Configuration overrides. Defaults to None.
    """
    self.args = get_cfg(cfg, overrides)
    self.callbacks = defaultdict(list, callbacks.default_callbacks)  # add callbacks
    callbacks.add_integration_callbacks(self)