Project procedurally generates trees and trains SP-GAN to generate them.
Refer to original README for model usage
Trees were procedurally generated with this blender addon
See the scripts in dataset_generation for dataset creation. generate_tree_dataset.py is a blender script to export the tree meshes and convert_dataset.py will convert them to pointclouds.
Two datasets were generated. One only uses the acer tree preset in the blender addon and the other uses all 20 preset types. They are available here
Two models were trained for 300 epochs. Some results are shown below. For training graphs and more examples see results. The pretrained models can be found here
