6D Pose Estimation with Synthetic Data and VR-Based Generation

DenseFusion-based assistive grasping pipeline with Unity/HTC Vive synthetic data generation.

This project focuses on 6D object pose estimation for assistive grasping. The workflow combines DenseFusion with a synthetic-data pipeline generated from Unity and HTC Vive captures, then evaluates pose estimation quality on a bottle-class setup.

Key outcomes:

  • Built synthetic data workflow: frame generation, mask extraction, and LineMOD-style formatting.
  • Integrated DenseFusion fine-tuning and evaluation on custom data.
  • Achieved 60.2% correct prediction rate on synthetic bottle-class evaluation.

Tech stack:

  • Python
  • DenseFusion
  • Unity
  • HTC Vive
  • Detectron2
  • MiDaS

Links: