Utilities for enabling easier high-level usage of TensorRT in Python.
The TRTEngine is a high-level abstraction allowing easy use of TensorRT engines through Python. Once an engine is built, it is simple and easy to use:
from trtutils import TRTEngine
engine = TRTEngine("path_to_engine")
inputs = read_your_data()
for i in inputs:
print(engine.execute(i))We also provide an abstraction for defining higher-level models. The TRTModel is designed to allow a user to define a pre and post processing step along with the engine to create an end-to-end inference object.
from trtutils import TRTModel
# scale some images down
def pre(inputs):
return [i / 255 for i in inputs]
# access the output classes from object detection
def post(outputs):
return [o[0][0] for o in outputs]
model = TRTModel("path_to_engine", pre, post)
inputs = read_your_data()
for i in inputs:
print(model(i))