o W hq"@sfddlZddlmZddlZddlmZmZddgdgdfdd Z          dd d ZdS)N)Path) IS_JETSONLOGGERimagesoutput0Fc Cs&tjj|||d|d|||pdd dS)at Exports a PyTorch model to ONNX format. Args: torch_model (torch.nn.Module): The PyTorch model to export. im (torch.Tensor): Example input tensor for the model. onnx_file (str): Path to save the exported ONNX file. opset (int): ONNX opset version to use for export. input_names (list): List of input tensor names. output_names (list): List of output tensor names. dynamic (bool | dict, optional): Whether to enable dynamic axes. Defaults to False. Notes: - Setting `do_constant_folding=True` may cause issues with DNN inference for torch>=1.12. FTN)verbose opset_versiondo_constant_folding input_names output_names dynamic_axes)torchonnxexport) torch_modelim onnx_fileopsetr r dynamicrL/var/www/vscode/kcb/lib/python3.10/site-packages/ultralytics/utils/export.py export_onnx s rrc  sddl|p t|d}jj} | rjjj| _| } | }t p)ddt j dddk}|rHdkrH| jjndkrO|_dt jj>}| || jo`|}| joe|}|durtsptdt| d |d |s|std jj|_t ||_|jj !| }|"|st#d |fd dt$j%D}fddt$j&D}|D]}t| d|j'd|j(d|j)q|D]}t| d|j'd|j(d|j)q|r6|ddkrt*| d| +}d|dddf}g|ddfdd|ddDR}|D] }|j,|j'|||dq#|-|t| d|r@dnd|rFdndd||r}|jj.|/|j0j1|_2Gfd d!d!j3}||t4t|d"d#|_5n |r|jj6|r| j7n| j8}||Q}t9|d$2}| durt:;| }|d%d&d'd(|<|?|<|r|n|@Wdn1swYWddSWddS1swYdS))a Exports a YOLO model to TensorRT engine format. Args: onnx_file (str): Path to the ONNX file to be converted. engine_file (str, optional): Path to save the generated TensorRT engine file. workspace (int, optional): Workspace size in GB for TensorRT. Defaults to None. half (bool, optional): Enable FP16 precision. Defaults to False. int8 (bool, optional): Enable INT8 precision. Defaults to False. dynamic (bool, optional): Enable dynamic input shapes. Defaults to False. shape (tuple, optional): Input shape (batch, channels, height, width). Defaults to (1, 3, 640, 640). dla (int, optional): DLA core to use (Jetson devices only). Defaults to None. dataset (ultralytics.data.build.InfiniteDataLoader, optional): Dataset for INT8 calibration. Defaults to None. metadata (dict, optional): Metadata to include in the engine file. Defaults to None. verbose (bool, optional): Enable verbose logging. Defaults to False. prefix (str, optional): Prefix for log messages. Defaults to "". Raises: ValueError: If DLA is enabled on non-Jetson devices or required precision is not set. RuntimeError: If the ONNX file cannot be parsed. Notes: - TensorRT version compatibility is handled for workspace size and engine building. - INT8 calibration requires a dataset and generates a calibration cache. - Metadata is serialized and written to the engine file if provided. rNz.enginei@. rz.DLA is only available on NVIDIA Jetson devicesz enabling DLA on core z...ztDLA requires either 'half=True' (FP16) or 'int8=True' (INT8) to be enabled. Please enable one of them and try again.zfailed to load ONNX file: cg|]}|qSr) get_input.0inetworkrr z!export_engine..cr r) get_outputr"r%rrr'r(z input "z " with shape z output "z> 'dynamic=True' model requires max batch size, i.e. 'batch=16' c3s&|]}ttdp d|VqdS)rN)intmax)r#d) workspacerr s$z export_engine..)minoptr.z building INT8FP1632z engine as csneZdZdZ ddeddffdd Zdjfdd Zdefd d Z de fd d Z de fddZ dddZdS)z'export_engine..EngineCalibratora  Custom INT8 calibrator for TensorRT. Args: dataset (object): Dataset for calibration. batch (int): Batch size for calibration. cache (str, optional): Path to save the calibration cache. Defaults to "". rcachereturnNcs<j|||_t||_jj|_|j|_ t ||_ dS)N) IInt8Calibrator__init__datasetiter data_iterCalibrationAlgoTypeENTROPY_CALIBRATION_2algo batch_sizebatchrr8)selfr<r8trtrrr;s   z0export_engine..EngineCalibrator.__init__cSs|jS)z%Get the calibration algorithm to use.)rArDrrr get_algorithmsz5export_engine..EngineCalibrator.get_algorithmcSs |jpdS)z*Get the batch size to use for calibration.r)rCrGrrrget_batch_sizes z6export_engine..EngineCalibrator.get_batch_sizecSsRzt|jdd}|jjdkr|dn|}t|gWSty(YdSw)zOGet the next batch to use for calibration, as a list of device memory pointers.imggo@cpucudaN)nextr>devicetypetor-data_ptr StopIteration)rDnamesim0srrr get_batchs z1export_engine..EngineCalibrator.get_batchcSs(|jr|jjdkr|jSdSdS)zSUse existing cache instead of calibrating again, otherwise, implicitly return None..cacheN)r8existssuffix read_bytesrGrrrread_calibration_caches z>export_engine..EngineCalibrator.read_calibration_cachecSs|j|}dS)z Write calibration cache to disk.N)r8 write_bytes)rDr8_rrrwrite_calibration_cachesz?export_engine..EngineCalibrator.write_calibration_cache)r)r9N)__name__ __module__ __qualname____doc__strr;r?rHr-rIlistrUbytesrZr]rrErrEngineCalibrators   rerV)r<r8wblittleT) byteordersigned)Atensorrtr with_suffixLoggerINFOSeverityVERBOSE min_severityBuildercreate_builder_configr- __version__splitset_memory_pool_limitMemoryPoolType WORKSPACEmax_workspace_sizeNetworkDefinitionCreationFlagEXPLICIT_BATCHcreate_networkplatform_has_fast_fp16platform_has_fast_int8r ValueErrorrinfo DeviceTypeDLAdefault_device_typeDLA_coreset_flag BuilderFlag GPU_FALLBACK OnnxParserparse_from_file RuntimeErrorrange num_inputs num_outputsnameshapedtypewarningcreate_optimization_profile set_shapeadd_optimization_profiler4set_calibration_profileProfilingVerbosityDETAILEDprofiling_verbosityr:rbint8_calibratorFP16build_serialized_network build_engineopenjsondumpswritelento_bytesencode serialize)r engine_filer0halfint8rrdlar<metadatarprefixloggerbuilderconfigis_trt10flagparserinputsoutputsinpoutprofile min_shape max_shaperebuildenginetmetar)r&rFr0r export_engine0s(          (( , 2  2   Tr) NNFFFrNNNFr) rpathlibrrultralytics.utilsrrrrrrrrs*  '