Responsibilities include:

• Worked on several applied research projects, including MediHopps, iperMö, FläKI and VuLCAn.
• Implemented advanced deep learning methods for human action recognition (HAR) and body pose estimation (HPE), and delivered detailed performance evaluations of these models, along with a trained HAR model (ST-GCN++) for custom rehabilitation exercise data captured in the lab using Meat Quest VR Headset.
• Systematic Literature Research and Reviews, Project Proposals and Scientific Literature Writing.
• Generally worked on computer vision, computer graphics, and machine/deep learning tasks like human pose estimation, human action recognition, and some XR tasks.

Responsibilities include:

• Worked on Radiance Field methods for Novel View Synthesis of structural color objects created by laser markings.
• Benchmarked SOTA radiance methods for synthetic scene involving Structural Color Object created in Blender.
• Developed capture setup to capture highly reflective and shiny structural color paintings on metal substrates.
• Improved the scene optimization using geometric prior and Anisotropy Regularizer in 3D Gaussian-Splatting method.
• Presented comprehensive experiments to demonstrate methods for simulating structural color objects before printing them using only captured images of laser-printed primaries.
• Facilitated interactive visualization of view dependent structural colors of laser-printed images and paintings on metal substrates in web viewer.

Responsibilities include:

• Developed dataset for adhesive test and corrosion detection on images of test panels of metal substrates.
• Developed framework and trained YOLOv8 model for adhesive tests’ detection and UNet for corrosion detection using created dataset for automation project.

Responsibilities include:

• Contributed to marker-less motion capture solutions using single and multiple cameras, for athlete motion tracking and analysis.
• Conducted a comprehensive literature research and review focused on deep learning approaches for human pose estimation and benchmarked SOTA approaches for domain specific video data.
• Worked on different tasks such as camera calibration, deep learning based human pose estimation & golf sequence detection, estimating joint angles from 3D body poses, comparing two pose sequences and visualization of results in Blender and Unity.