tfm.vision.serving.export_saved_model_lib.export_inference_graph

Exports inference graph for the model specified in the exp config.

tfm.vision.serving.export_saved_model_lib.export_inference_graph(
    input_type: str,
    batch_size: Optional[int],
    input_image_size: List[int],
    params: tfm.core.base_trainer.ExperimentConfig,
    checkpoint_path: str,
    export_dir: str,
    num_channels: Optional[int] = 3,
    export_module: Optional[tfm.core.export_base.ExportModule] = None,
    export_checkpoint_subdir: Optional[str] = None,
    export_saved_model_subdir: Optional[str] = None,
    save_options: Optional[tf.saved_model.SaveOptions] = None,
    log_model_flops_and_params: bool = False,
    checkpoint: Optional[tf.train.Checkpoint] = None,
    input_name: Optional[str] = None,
    function_keys: Optional[Union[List[Text], Dict[Text, Text]]] = None,
    add_tpu_function_alias: Optional[bool] = False
)

Saved model is stored at export_dir/saved_model, checkpoint is saved at export_dir/checkpoint, and params is saved at export_dir/params.yaml.

input_type One of image_tensor, image_bytes, tf_example or tflite.
batch_size 'int', or None.
input_image_size List or Tuple of height and width.
params Experiment params.
checkpoint_path Trained checkpoint path or directory.
export_dir Export directory path.
num_channels The number of input image channels.
export_module Optional export module to be used instead of using params to create one. If None, the params will be used to create an export module.
export_checkpoint_subdir Optional subdirectory under export_dir to store checkpoint.
export_saved_model_subdir Optional subdirectory under export_dir to store saved model.
save_options SaveOptions for tf.saved_model.save.
log_model_flops_and_params If True, writes model FLOPs to model_flops.txt and model parameters to model_params.txt.
checkpoint An optional tf.train.Checkpoint. If provided, the export module will use it to read the weights.
input_name The input tensor name, default at None which produces input tensor name inputs.
function_keys a list of string keys to retrieve pre-defined serving signatures. The signaute keys will be set with defaults. If a dictionary is provided, the values will be used as signature keys.
add_tpu_function_alias Whether to add TPU function alias so that it can be converted to a TPU compatible saved model later. Default is False.