Atharva Vaidya

• Developed real-time multiview RGB and RGBD-based 3D human pose estimation pipelines • Real-time surgical tool tracking and motion capture of human hands • Generated high-fidelity synthetic data for instance segmentation • ControlNet-based texture synthesis • Digital twin toolbox based on OpenUSD • Conversational language understanding model by Azure AI Language Studio

Human pose estimation Generating synthetic data with automatic annotations Domain randomization in the synthetic data for reduction in domain gap Training pipelines of CNN and Transformer models like YOLOv7, YOLOv8, ViTPose for synthetic-to-real transfer learning Domain adaptation in YOLOv7 architecture Triangulated the 2D keypoint predictions to 3D using multi-view images

Involved in the development of a LDR based optical sensor used for identification of impurities dissolved in water

Graphene synthesis using Chemical Vapor Deposition

Defect reduction in Fuel-tank Weld line