Installation

Follow the steps below to setup PV Hawk on your machine.

Step 1: Fullfill prerequisites

PV Hawk requires a machine running a 64-bit version of Ubuntu 21.10, 21.04, 20.04 LTS or 18.04 LTS with Docker CE installed.

To make use of GPU acceleration (highly recommended) you need a Nvidia CUDA-compatible GPU with the latest Nvidia drivers and you must install the Nvidia container toolkit.

Step 2: Download source code

Open a new terminal and navigate to the location where you want to install PV Hawk, e.g. /software. Run the command below to clone the Git repository to your machine

git clone https://github.com/LukasBommes/PV-Hawk

Step 3: Download Mask R-CNN weights

PV Hawk uses a pretrained Mask R-CNN for PV module detection. We provide pretrained models for thermal IR and RGB videos. Download the pretrained IR model from here and the RGB model from here. Place both model files under extractor/segmentation/Mask_RCNN.

Step 4: Pull Docker image

We provide a prebuilt Docker image containing all runtime dependencies of PV Hawk, such as Python, CUDA and Tensorflow. All you have to do is pull the Docker image with

sudo docker pull lubo1994/pv-hawk:latest

If you run into problems with the prebuilt image, you can instead Build the Docker Image locally.

Step 5: Test the installation

PV Hawk comes with some test cases, which you can run to test whether the installation was successful. To this end, start a terminal in the source code root directory and run the following two commands

xhost +

sudo docker run -it \
  --ipc=host \
  --env="DISPLAY" \
  --gpus=all \
  --mount type=bind,src=/tmp/.X11-unix,dst=/tmp/.X11-unix \
  --mount type=bind,src="$(pwd)",dst=/pvextractor \
  --mount type=volume,dst=/pvextractor/extractor/mapping/OpenSfM \
  lubo1994/pv-hawk:latest \
  bash

This should start an interactive shell in the Docker container. Run the tests in that shell with

python -m unittest tests/**/test_*.py

If everything was installed correctly you should see an OK message after a few minutes. If you see any failures, confirm that you followed the steps above correctly. You can also open an issue in the GitHub repository.