This code implements a full ONNX-based pipeline for performing inference with the YOLOv3 network, using int8 calibration. This code is partly based on the offical sample "yolov3_onnx.py" given by Tensor RT. This sample is based on the YOLOv3-608 paper.
Note: This sample is not supported on Ubuntu 14.04 and older. Additionally, the yolov3_to_onnx.py script does not support Python 3.
For specific software versions, see the TensorRT Installation Guide.
-
Install ONNX-TensorRT: TensorRT backend for ONNX. ONNX-TensorRT includes layer implementations for the required ONNX operators
UpsampleandLeakyReLU. -
Install the dependencies for Python.
-
For Python 2 users, from the root directory, run:
python2 -m pip install -r requirements.txt -
For Python 3 users, from the root directory, run:
python3 -m pip install -r requirements.txt
-
-
Create an ONNX version of YOLOv3 with the following command. The Python script will also download all necessary files from the official mirrors (only once).
python yolov3_to_onnx.pyWhen running this sample you could get middle layer output by changing the definition of variable "output_layer_name", layer name can be get in yolov3.onnx
-
Build a TensorRT engine from the generated ONNX file and run inference on a sample image, which will also be downloaded during the first run.
python onnx_to_tensorrt_int8.pyWhen running this sample you could get middle layer output by changing the definition of variable "output_layer_name", layer name can be found in yolov3.onnx
-
Verify that the sample ran successfully. If the sample runs successfully you should see output similar to the following:
The following resources provide a deeper understanding about the model used in this sample, as well as the dataset it was trained on:
Model
Dataset
Documentation