How to Reconstruct Plants with the Geometric Pipeline

In this step, we reconstruct the 3D structure of the plant and quantify its phyllotaxy.

Getting Started

1. Open a terminal by pressing Ctrl + Alt + t.

2. Activate the plant3dvision conda environment (A) OR start a Docker container (B):

   1. Conda environment:

      conda activate plant3dvision
      

   2. Docker container:

      bash ${HOME}/Projects/plant-3d-vision/docker/run.sh \
          -db /data/ROMI/test_v2 \
          -v /data/ROMI/configs:/myapp/configs \
          -v ~/scripts:/myapp/scripts \
          -t latest \
          -c /bin/bash
      

3. Copy your acquisition data, for example /data/ROMI/Romi_****/acquisitions/my_awesome_plant_*, to a different directory for analysis, such as /data/ROMI/Romi_****/analysis:

   cp /data/ROMI/Romi_Fabfab/acquisitions/my_awesome_plant_* /data/ROMI/Romi_Fabfab/analysis/.
   

Reconstruction

To proceed with the reconstruction of a new plant, for example, named my_awesome_plant_007, follow these steps:

1. Verify the parameters in the TOML configuration file, for example /data/ROMI/configs/pipeline.toml:

   [Colmap]
   upstream_task = "ImagesFilesetExists"
   matcher = "exhaustive"
   align_pcd = true
   use_gpu = true  # faster processing 
   single_camera = true  # images all comes from a single camera
   distance_threshold = 5  # in mm; maximum allowed Euclidean distance from CNC pose
   
   [Undistort]
   upstream_task = "ImagesFilesetExists"
   query = "{\"channel\":\"rgb\", \"pose_estimation\":\"correct\"}"  # RGB images with a valid COLMAP pose
   
   [Masks]
   upstream_task = "Undistort"
   type = "linear"
   parameters = "[0.2, 1, 0.1]"
   dilation = 3 #Modified
   binarize = true
   threshold = 0.2
   
   [Voxels]
   upstream_mask = "Masks"
   upstream_colmap = "Colmap"
   voxel_size = 0.6
   type = "averaging"
   [Voxels.bounding_box]
   x = [ 270, 465,]
   y = [ 270, 465,]
   z = [ -320, 50,]
   
   [PointCloud]
   upstream_task = "Voxels"
   level_set_value = 0.0
   
   [TriangleMesh]
   upstream_task = "PointCloud"
   library = "open3d"
   filtering = "largest connected triangles"
   depth = 9
   
   [Clean]
   no_confirm = true
   

2. Start the reconstruction with:

   romi_run_task TriangleMesh /data/ROMI/Romi_Fabfab/analysis/my_awesome_plant_007 --config /data/ROMI/configs/pipeline.toml