Tree-GAN

Implementation of Tree-GAN paper.

Dataset Generation Step

• ModelNet10 dataset is used.
• To sample pointcloud from mesh:
  • https://medium.com/@daviddelaiglesiacastro/3d-point-cloud-generation-from-3d-triangular-mesh-bbb602ecf238
  • Area of triangle with 3 coordinates: https://math.stackexchange.com/a/1951650
  • https://www.youtube.com/watch?v=HYAgJN3x4GA
• Data generation code is present in Point_Sampling_From_Mesh folder.
• ModelNet10 numpy format dataset: Link

Observations / Notes

• Keeping larger values for batch size helps model in learning better distribution for pointcloud data.

Pre-trained model

• Download pre-trained model from googl drive: Link
• Keep treegan_ckpt folder as it is in code directory.

Results

Todo Task

• Explaination of the data generation part.
• Add results.

Reference

[1] 3D Point Cloud Generative Adversarial Network Based on Tree Structured Graph Convolutions [ Dong Wook Shu, Sung Woo Park, Junseok Kwon ]