Projects
Reverse Pass-through VR framework
Description:
- Developed a general framework for reverse pass-thorugh VR.
- Implemented eye image aligment, reference based eye colorization models for this framework.
GANxNeRF: GAN based universal NeRF output refinement model framework
Description:
- Developed a multiscale conditional GAN model to refine the outputs of any NeRF based models.
- The GAN model is NeRF model agnostic and can recover details which are generally not captured by NeRF based models.
POC of Apple’s EyeSight features for VR headsets, Samsung Research America
Description:
- Implemented the POC for Apple’s EyeSight feature.
- My approach does not require any 3D face scan to be taken by the user. This greatly simplifies the overall experience of using VR headset.
- My approach produces much richer facial expression and looks more natural
Output form my approach
Apple's EyeSight Demo
HI-GAN: RGBD inpainting using Hierarchical GAN and auxiliary inputs
Description:
- Developed HI-GAN model for RGB-D inpaiting.
- Hierarchical-GAN, a novel approach comprising three GANs for RGBD inpainting.
- EdgeGAN and LabelGAN process edge and segmentation label images respectively, while CombinedGAN combines their latent representation outputs and performs RGB and Depth inpainting.
Computer Vision for AR/VR application, Samsung Research America
Description:
- Developed Deep Learning based Computer Vision models for improving AR/VR user experience.
- Successfully carried out image and video inpainting methods for improving LSR output for AR/VR applications.
- The developed model can perform combined RGBD inpainting.
High Resolution Solar Image Generation using Generative Adversarial Networks
Description:
- Developed Deep Learning models to generate Solar images using Image-to-Image translation.