Jan 2020 – Jul 2021 · 1 yr 6 mos · Baltimore, Maryland, United States

• 1. Graduate Research Assistant in Vision Lab under Prof. Rene Vidal
• 2. Zero Shot Learning for Image Classification :
• ∗ Researched on a BiGAN framework to derive disentangled representations for images and videos, that allows zero-shot classification of novel categories using only semantic descriptions and no visual data.
• ∗ Achieved near state-of-the-art results on SUN and UCF101 datasets.
• 3. Activity Detection from Surveillance Videos:
• ∗ Built a dockerized multi-GPU app to detect human activities in surveillance videos.
• ∗ Involved tracking people and processing videos with deep networks like Faster-RCNN and I3D models.
• ∗ Achieved real-time video processing and recorded area under DET curve of 0.64

Aug 2018 – Feb 2019 · 6 mos · Mumbai

• 1. Generative Model for 3D Human Pose Estimation:
• ∗ Researched on an architecture to predict 2D human pose from monocular images and generate 3D pose
• hypotheses with a Conditional Variational Autoencoder. Published in ICCV’19 [Paper]
• ∗ Achieved near state-of-the-art results on Human3.6M and HumanEva datasets.
• 2. View-Invariant 3D Human Pose Estimation with Metric Learning:
• ∗ Researched on an architecture to predict 2D human pose from monocular images and generate 3D pose
• hypotheses with a Conditional Variational Autoencoder. Published in ICCV’19 [Paper]
• ∗ Achieved near state-of-the-art results on Human3.6M and HumanEva datasets.

Aug 2015 – Jul 2019 · 3 yrs 11 mos · Bengaluru Area, India

• Incremental Data Processing Platform:
• ∗ Developed a pipeline for incremental replication of data from relational databases for creating a
• denormalized analytical data model in Elasticsearch.
• ∗ Staged the relational tables in Vertica to perform multi-way joins across hundreds of tables.
• ∗ Created a service to generate Vertica join query on-the-fly by modeling tables and their join relationships as graph nodes and edges respectively.
• ∗ Implemented a Spark application to enrich data from 3rd party APIs with generic user-defined functions.
• ∗ Achieved data freshness SLA of under 10 minutes at the scale of 5000 events/sec.
• Logistics Analytics Platform:
• ∗ Built a Spark streaming application to process supply chain data from Kafka and persist to Elasticsearch.
• ∗ Implemented a state machine in Redis for tracking shipment status and handling out-of-order events.
• ∗ Developed the full stack for multiple RESTful webapps to serve the processed data on analytical dashboards for end-to-end supply chain visibility to the company leadership

Jan 2015 – Jun 2015 · 5 mos · Columbus, Ohio Area

• Augmented Reality Assisted Cancer Surgeries:
• ∗ Created a prototype to display infrared fluorescence of cancer tumors on Google Glass in augmented reality.
• ∗ Involved registering images from near-infrared camera with Microsoft Kinect to get 3D coordinates of tumor
• and displaying augmented reality output on Google Glass with perspective projection.

May 2013 – Jul 2013 · 2 mos · New Delhi Area, India

• Educational Video Content Analysis:
• ∗ Built a pipeline to index videos with keyframe detection by analyzing connected components and edge maps
• of video frames.
• ∗ Parsed text on keyframes with Tesseract OCR and detected keywords using tf-idf score.

May 2013 – Jul 2013 · 2 mos · Dehradun

• ∗ Built a pipeline to index videos with keyframe detection by analyzing connected components and edge maps
• of video frames.
• ∗ Parsed text on keyframes with Tesseract OCR and detected keywords using tf-idf score.