YOLOv2 (You only look once) is a widely used in computer vision tasks such as face detection, object recognition, etc. This is a Tensorflow 2.1.0 implementation of the same architecture (with slight modifications) for traffic assesment.
The Jupyter notebook could be viewed in Google Collab via the link here.
First, we need to setup the working enviourment for the project. A simple way of doing so is Another relateivly easy way to get started is to install Docker and the official Tensorflow image.
Which image you should install primairly depends on whether you have NVIDIA Cuda drivers installed. If so, you will need to install nvidia-docker and you will be able to run the GPU supported tensorflow/tensorflow:nightly-gpu-py3-jupyter image. Otherwise you can still run the CPU-only image - tensorflow/tensorflow:nightly-py3-jupyter.
For CPU-only supporting image run the following:
$ docker run -it -v $(pwd):/tf -p 8888:8888 tensorflow/tensorflow:nightly-py3-jupyter
For GPU supporting image run the following:
$ docker run -it --runtime=nvidia -v $(pwd):/tf -p 8888:8888 tensorflow/tensorflow:nightly-gpu-py3-jupyter
The notebook should be accessible on localhost:8888 with the access token printed in the command-prompt.
You can a much more detailed explanation on YOLOv1 in the research paper - here and on YOLOv2 - here.
Python 3.7 with the following libraries:
- Tensorflow (2.1.0) - Neural network model
- Numpy (1.18.2) - Data formating
- Pillow - Loading images
- Jupyter Notebook (1.0.0) - IDE
- MatPlotLib (3.2.1) - Data visualization
Distributed under the MIT license. See LICENSE for more information.