Project information

  • Client: Changins (University of viticulture and oenology) / HE-Arc
  • Categories: Computer Vision, 3D Point Clouds, Deep Learning
  • Main technologies: Python, PyTorch, Point Cloud Processing, Image Masking

Summary

Estigrappe3D was developed in collaboration with Changins, the university of viticulture and oenology, to automate the weight estimation of grapes using depth imaging and 3D point clouds.

Working with a dataset of depth-aware images, I built a specialized computer vision pipeline. First, I integrated a masking model to accurately segment the grape bunches and isolate them from complex background environments.

Using the extracted 3D point cloud data, I designed and trained a custom PyTorch deep learning model capable of accurately predicting the physical weight of the grapes directly from their spatial representation.