Jul 2021 – Jul 2022 · 1 yr · Noida, Uttar Pradesh, India

• Worked on storage and business services of Acrobat application.

Jan 2021 – Jun 2021 · 5 mos · Noida, Uttar Pradesh, India

• Worked with Media and Data Science Research Team. Developed mitigation strategies against two biases in existing one-shot segementation methods. The work got accepted in NeurIPS 2021.

May 2020 – Jul 2020 · 2 mos · Noida, Uttar Pradesh, India

• Worked on Few-Shot Image Segmentation as part of the Media and Data Science Research Team.
• Developed end-to-end deep learning based system to address broadly two main problems - 1) issue with the training images having test class objects; 2) inability of networks to attend to non-salient objects and reported results on public datasets Pascal-5i and COCO-20i improving upon SOTA methods by ~2% (in mIoU). The work got accepted in Adobe Tech Summit 2021 - a biyearly internal conference held across Adobe.

May 2019 – Aug 2019 · 3 mos

• Gesture Detection and Recognition in TV News Videos
• Aim of the project was to develop an automated system to detect the presence of hand gestures, segment them temporally and further classifiy the detected hand gestures. Worked on NewsScape Dataset - Red Hen Lab's news video dataset. Also, used singularity container for code deployment on high performance clusters.

May 2019 – Jul 2019 · 2 mos · Bangalore, India

• Generic HBase Plug-in Implementation
• Worked as a part of Global API Platform of Data Product Development Team
• Developed Java API wrapper over HBase Native API (HBase is a Database on Hadoop Distributed File System)
• Integrated this HBase Plugin Microservice with the Global API Platform
• Got exposure of HBase architecture and working of HBase Native API

May 2018 – Jul 2018 · 2 mos · Roorkee, Uttarakhand

• Single-Image based Rain and Haze Detection and Removal
• Worked under - Dr. Sanjeev Kumar (Assistant Professor – Department of Mathematics, IIT Roorkee). Proposed a method for removal of rain and haze from images captured in bad weather conditions (rain, haze, fog etc). The method is a combination of CNN model and Dark Channel Prior technique for removing rain and haze respectively.