Nov 2020 – Present · 5 yrs 4 mos · Mountain View, California, United States

• Generating talking faces from audio

May 2019 – Aug 2019 · 3 mos · Mountain View, California

• 3D face model generation from audio
• Working at the intersection of speech and video with VAE and GAN.

Jun 2018 – Sep 2018 · 3 mos · San Francisco Bay Area

• Mobile MaskRCNN for real-time human pose estimation from videos with special concern for mobile deployment use cases.
• 1. Experience with Caffe2
• 2. Pose Estimation, Object Detection, Instance Segmentation
• 3. Mobile friendly deep learning architectures
• 4. Video Processing for Pose Estimation

Jun 2017 – Sep 2017 · 3 mos · London, United Kingdom

• super resolution in videos using GANs
• sequence consistency learning
• temporal GAN

May 2016 – Aug 2016 · 3 mos · Mountain View, USA

• Designed complex deep neural network architectures for multiscale processing of videos to predict depth from RGB frames
• Had to design specialized networks to reduce error of depth predictions on video conferencing sequences. Proposed design significantly outperforms latest state-of-the-art

May 2015 – Jul 2015 · 2 mos · Los Angeles, USA

• During this internship I worked under Prof. Ram Nevatia on Fisher Vector encoding of dense motion trajectories for action recognition/localization on large scale YouTube video datasets.

Dec 2013 – Dec 2015 · 2 yrs · Kharagpur Area, India

• Collaborative project with IIT Kharagpur and ISRO for my MS thesis.
• I was responsible for designing algorithms for video error concealment in video-conferencing scenario. I have proposed a deformable wiremesh face-model which enables to conceal dropped packets on facial region. In this regard I also worked on collaborative adaptive boosting algorithms for detecting facial landmarks. The proposed boosting platform is generic and can be used for object/speaker recognition, biometric, etc.

Jun 2012 – Nov 2012 · 5 mos

• I was responsible for writing efficient data transfer APIs using C++.