NvInferModel

class savant.deepstream.nvinfer.model.NvInferModel(local_path=None, remote=None, model_file=None, batch_size=1, precision=ModelPrecision.FP16, input=NvInferModelInput(object='auto.frame', layer_name=None, shape=None, maintain_aspect_ratio=False, symmetric_padding=False, scale_factor=1.0, offsets=(0.0, 0.0, 0.0), color_format=<ModelColorFormat.RGB: 0>, preprocess_object_meta=None, preprocess_object_image=None, object_min_width=None, object_min_height=None, object_max_width=None, object_max_height=None), format=None, config_file=None, int8_calib_file=None, engine_file=None, proto_file=None, custom_config_file=None, mean_file=None, label_file=None, tlt_model_key=None, gpu_id=0, interval=0, workspace_size=6144, custom_lib_path=None, engine_create_func_name=None, layer_device_precision=<factory>, enable_dla=None, use_dla_core=None, scaling_compute_hw=None, scaling_filter=None)

Base configuration template for a nvinfer model.

input: NvInferModelInput = NvInferModelInput(object='auto.frame', layer_name=None, shape=None, maintain_aspect_ratio=False, symmetric_padding=False, scale_factor=1.0, offsets=(0.0, 0.0, 0.0), color_format=<ModelColorFormat.RGB: 0>, preprocess_object_meta=None, preprocess_object_image=None, object_min_width=None, object_min_height=None, object_max_width=None, object_max_height=None): Optional configuration of input data and custom preprocessing methods for a model. If not set, then input will default to entire frame.

format: NvInferModelFormat | None = None

Model file format.

Example

format: onnx
# format: caffe
# etc.
# look in enum for full list of format options

config_file: str | None = None: Model configuration file. Should be specified without a location.

int8_calib_file: str | None = None: INT8 calibration file for dynamic range adjustment with an FP32 model. Required only for models in INT8.

batch_size: int = 1: Number of frames or objects to be inferred together in a batch.

Note

In case the model is an NvInferModel and it is configured to use the TRT engine file directly, the default value for batch_size will be taken from the engine file name, by parsing it according to the scheme {model_name}_b{batch_size}_gpu{gpu_id}_{precision}.engine

engine_file: str | None = None: Serialized model engine file.

local_path: str | None = None: Path where all the necessary model files are placed. By default, the value of module parameter “model_path” and element name will be used (“model_path / element_name”).

model_file: str | None = None: The model file, eg yolov4.onnx.

Note

The model file is specified without a location. The absolute path to the model file will be defined as “local_path/model_file”.

precision: ModelPrecision = 2

Data format to be used by inference.

Example

precision: fp16
# precision: int8
# precision: fp32

Note

In case the model is an NvInferModel and it is configured to use the TRT engine file directly, the default value for precision will be taken from the engine file name, by parsing it according to the scheme {model_name}_b{batch_size}_gpu{gpu_id}_{precision}.engine

remote: RemoteFile | None = None: Configuration of model files remote location. Supported schemes: s3, http, https, ftp.

proto_file: str | None = None: Caffe model prototxt file. By default, the model file name (model_file) will be used with the extension .prototxt.

custom_config_file: str | None = None: Configuration file for custom model, eg for YOLO. By default, the model file name (model_file) will be used with the extension .cfg.

mean_file: str | None = None: Pathname of mean data file in PPM format.

label_file: str | None = None: Pathname of a text file containing the labels for the model.

tlt_model_key: str | None = None: Key for the TAO toolkit encoded model.

gpu_id: int = 0: Device ID of GPU to use for pre-processing/inference (dGPU only).

Note

In case the model is configured to use the TRT engine file directly, the default value for gpu_id will be taken from the engine_file, by parsing it according to the scheme {model_name}_b{batch_size}_gpu{gpu_id}_{precision}.engine

interval: int = 0: Specifies the number of consecutive batches to be skipped for inference.

workspace_size: int = 6144: Workspace size to be used by the engine, in MB

custom_lib_path: str | None = None: Absolute pathname of a library containing custom method implementations for custom models.

engine_create_func_name: str | None = None: Name of the custom TensorRT CudaEngine creation function.

layer_device_precision: List[str]: Specifies the device type and precision for any layer in the network. List of items of format <layer1-name>:<precision>:<device-type>.

enable_dla: bool | None = None: Specifies that DLA engines to be used for inference.

use_dla_core: int | None = None: Specifies the DLA core to be used for inference.

scaling_compute_hw: NvInferScalingComputeHW | None = None: Specifies the hardware to be used for scaling compute.

scaling_filter: NvInferScalingFilter | None = None: Specifies the algorithm to be used for scaling.