CropCraft is a python script that generates 3D models of crop fields, specialized in real-time simulation of robotics applications.

• Designed for real-time simulation
• Suitable for use with LiDARs and cameras
• Highly configurable (YAML file)
• Provide ground truth data (identify plant types in LiDAR data)

This program uses blender as a backend. It is a 3D modeling software that you can dowload from the official website. If you use Ubuntu, you can install it using snap:

snap install blender --classic 

The minimal required version is 4.0. Ensure that blender is launchable from the command line. It means that blender must be accessible using the PATH environment variable.

You also need to install some python requirements:

pip install -r requirements.txt 

To generate a crop field, you first need to create a configuration file (YAML formats). Some examples are available in the examples directory. Then you can execute the cropcraft.py script and specify the path of the chosen configuration file.

python cropcraft.py examples/test1.yaml 

This command will generate a blender file named test1.blend and a gazebo model named test1

Some options are available and described using

python cropcraft.py --help 

CropCraft can render a dataset of paired RGB images and semantic segmentation masks by adding a render block to your configuration file:

render: directory: render # output subdirectory (relative to --output-dir) frames: 1 # maximum number of frames to render samples: 32 # number of CYCLES samples per pixel (RGB pass) camera: height: 1.6 # camera height above the ground (meters) fov_deg: 60.0 # horizontal field of view (degrees) roll_deg: 0.0 # X-axis rotation in degrees (XYZ Euler order) pitch_deg: 0.0 # Y-axis rotation in degrees yaw_deg: 0.0 # Z-axis rotation in degrees y_jitter: 0.1 # random lateral offset per frame (meters) label_colors: # RGB colors for semantic mask classes (optional) crop: [0, 255, 0] weed: [255, 0, 0] background: [0, 0, 0]

When a render block is present, CropCraft renders two passes in a single run:

1. RGB images — rendered with Cycles, saved as JPEG in <directory>/images/
2. Semantic masks — rendered with EEVEE using flat emission materials, saved as PNG in <directory>/masks/
   • Crops → label_colors.crop (default: green [0, 255, 0])
   • Weeds → label_colors.weed (default: red [255, 0, 0])
   • Ground / stones → label_colors.background (default: black [0, 0, 0])

Note: when using cycles_device: GPU, the desired GPU must be enabled in Blender beforehand. Open Blender and go to Edit > Preferences > System > Cycles Render Devices, then select your GPU. This setting persists across runs. (Verified on Blender 5.0.1.)

See examples/render_example.yaml for a complete example.

• Description of the configuration file format
• How to add your own assets