Information

• YOLOv3-TensorRT has benn deprecated. Please see tensorrt_inference/project/Yolov4, which has updated new YOLO models includes YOLOv3-SPP, YOLOv3-tiny, YOLOv4, YOLOv4-tiny, YOLOv4x-mish an so on.

Table Of Contents

• Introduction
• Prerequisites
• How does this sample work?
• Running the sample
• Reference

This sample, implements a full ONNX-based pipeline for performing inference with the YOLOv3 network, which the input size can be assigned by set --width and --height in onnx_to_tensorrt.py, and int8 mode can also be set by add --int8 in onnx_to_tensorrt.py, including pre and post-processing. This sample is based on the YOLOv3-608 paper.

For specific software versions, see the TensorRT Installation Guide.

1. Install ONNX-TensorRT: TensorRT backend for ONNX. ONNX-TensorRT includes layer implementations for the required ONNX operators Upsample and LeakyReLU.

2. 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

First, the original YOLOv3 specification from the paper is converted to the Open Neural Network Exchange (ONNX) format in yolov3_to_onnx.py (only has to be done once).

Second, this ONNX representation of YOLOv3 is used to build a TensorRT engine, followed by inference on a sample image in onnx_to_tensorrt.py. The predicted bounding boxes are finally drawn to the original input image and saved to disk.

After inference, post-processing including bounding-box clustering is applied. The resulting bounding boxes are eventually drawn to a new image file and stored on disk for inspection.

• Create an ONNX version of YOLOv3 with the following command.
  

   python3 yolov3_to_onnx.py --cfg_file yolov3.cfg --weights_file yolov3.weights --output_file yolov3.onnx
  

• Build a TensorRT engine from the generated ONNX file and run inference on a sample image.

   python3 onnx_to_tensorrt.py --onnx_file yolov3.onnx --engine_file yolov3.trt
  

  If you want to run in int8 mode, you can run the code like this

   python3 onnx_to_tensorrt.py --int8 --onnx_file yolov3.onnx --engine_file yolov3_int8.trt
  

• Saved image with bounding boxes of detected objects to dog_bboxes.png. You can verify that the sample ran successfully.
  

The following resources provide a deeper understanding about the model used in this sample, as well as the dataset it was trained on:

Dataset

• COCO dataset
• PASCAL VOC dataset

Documents

• YOLOv3: An Incremental Improvement
• Introduction To NVIDIA’s TensorRT Samples
• Working With TensorRT Using The Python API
• NVIDIA’s TensorRT Documentation Library