All projects

Human Avatars

Our 3D computer vision pipelines enable high-fidelity human capture and precise anthropometric measurement. By leveraging the SMPL model, we generate synthetic datasets featuring diverse body dimensions paired with accurate, ground-truth metadata for robust model training.

Generating synthetic datasets of human.

Avatar Generation

Our approach to developing metrically accurate avatars involves estimating anthropometric measurements from silhouette images to generate body model parameters while utilizing diffusion models to improve the quality of the synthetic data used for training and real-world evaluation.

Improving the quality of synthetic data using diffusion models with SMPL sillhouetes.

Realistic Head Rendering

Avatar rendering utilizes extraction of head model parameters from monocular video, anchoring Gaussian primitives to the centers of mesh triangles so that they remain locally static during animation and maintain their triangle assignments throughout adaptive density control.

Head scanning and rendering using Gaussian Splatting

Dental Scanning

By leveraging existing motion capture and pose estimation frameworks to animate randomly generated human models, our approach focuses on producing specialized synthetic datasets for critical computer vision tasks — such as detecting suspicious or dangerous activities.

We also specialize in the processing of dental data from 3D scans and 2D photographs, with capabilities ranging from 6D pose estimation for individual tooth localization and instance segmentation to the generation of synthetic training datasets.
All projects