About This Sample

This Python sample demonstrates how to run YOLOv3-608[^1] (with an input size of 608x608 pixels - in the following just referred to as YOLOv3) in TensorRT 5.0, using ONNX-TensorRT (https://github.com/onnx/onnx-tensorrt).

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
in onnx_to_tensorrt.py, followed by inference on a sample image. The predicted bounding boxes are finally drawn to the original input image and saved to disk.

Note that this sample is not supported on Ubuntu 14.04 and older. Additionally, the yolov3_to_onnx.py does not support Python 3.

Installing Prerequisites

1. Make sure you have the python dependencies installed.
   • For python2, run python2 -m pip install -r requirements.txt from the root directory of this sample.
   • For python3, run python3 -m pip install -r requirements.txt from the root directory of this sample.

Running the Sample

1. 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.py
   

   When running above command for the first time, the output should look like this:

   Downloading from https://raw.githubusercontent.com/pjreddie/darknet/
   f86901f6177dfc6116360a13cc06ab680e0c86b0/cfg/yolov3.cfg,
   this may take a while...
   100% [......................................................................
   ..........] 8342 / 8342
   Downloading from https://pjreddie.com/media/files/yolov3.weights, this may
   take a while...
   100% [......................................................................
   ] 248007048 / 248007048
   [...]
     %106_convolutional = Conv[auto_pad = u'SAME_LOWER', dilations = [1, 1],
     kernel_shape = [1, 1], strides = [1, 1]](%105_convolutional_lrelu,
     	%106_convolutional_conv_weights, %106_convolutional_conv_bias)
     return %082_convolutional, %094_convolutional, %106_convolutional
   }
   

2. After completing the first step, you can proceed to building a TensorRT engine from the generated ONNX file and running inference on a sample image (will also be downloaded during the first run). Run the second script with:

   python onnx_to_tensorrt.py
   

   Doing this for the first time should produce the following output:

   Downloading from https://github.com/pjreddie/darknet/raw/f86901f6177dfc6116
   360a13cc06ab680e0c86b0/data/dog.jpg, this may take a while...
   100% [.....................................................................
   .......] 163759 / 163759
   Building an engine from file yolov3.onnx, this may take a while...
   Running inference on image dog.jpg...
   Saved image with bounding boxes of detected objects to dog_bboxes.jpg.
   

References

[^1]: Redmon, Joseph, and Ali Farhadi. "Yolov3: An incremental improvement." arXiv preprint arXiv:1804.02767 (2018).