phytooracle/3d_landmark_selection repository overview

⁠3D Landmark Selection GUI

This repo is used to geo-correct the aligned 3D point clouds generated by the laser scanner mounted on the Gantry which is located at Maricopa Agricultural Center for University of Arizona. This code works in parallel with another repo (Phytooracle Data Repo). This repo is tested on the data generated for seasons 10, 11, and 12 so far and it works without any issue on those seasons. You can run the code using either Docker or this repo.

⁠Running landmark selection using container

For help running the container, click here⁠.

⁠Using the GUI

Run the following command from the 3d_landmark_selection folder in your terminal after activating the conda environment.

python main.py -s [date] -S [season_number] [-a]

[date] specifies the scan date you wish to geo-correct, [season_number] specifies the number of the season and -a is an optional argument specifying whether to use aligned or pre-processed outputs of the previous step.

After running this command, if your environment is correctly configured, the required data will be automatically downloaded from cyverse by use of iCommands. If the data is already downloaded, this step would be skipped. Once everything is downloaded successfully, the following window containing an RGB ortho will be visualized.

You can pan around this ortho and select landmarks by double clicking on a pixel. Note that once you do that, it would lock the current window and find nearby 3d point cloud files. Once those point clouds are merged together, another window will be visualized containing those point clouds. To mark a landmark on the RGB ortho with a blue circle, press your scroll button on your mouse.

In this window you can pan around the point cloud in 3d and select the corresponding point to the selected landmark. Use the following instructions and shortcut keys to work with this visualizer:

Once you selected enough landmarks in RGB and their corresponding point in the point cloud, press q in the ortho window.

We recommend selecting at leaset 5-10 landmark points, preferably scattered throughout the entire field. If enough landmarks were selected, a transformation is going to be calculated and if the transformation error is below a threshold, a json file containing the transformation will be created and uploaded to cyverse after the transformation is printed out in the terminal.