Projects

Semantic Segmentation and MonoDepth Estimation of SUADD23 Drone Images

The project's goal is to foster the development of fully autonomous Unmanned Aircraft Systems (UAS) by providing a benchmark dataset and challenge for two key computer vision components: semantic segmentation and depth perception.The dataset comprises realistic backyard scenarios of variable content taken on various Above Ground Level (AGL) ranges. The challenge aims to inspire the Computer Vision community to develop new insights and advance state-of-the-art in perception tasks involving drone images.

Java Code Generator-2021

Finetuned ethirium-6B model to generate Java code based on the input text. For data generation I have extracted top java GitHub repositories using google TPU research cloud.

Finetuned this model public git repositories and infosys's internal git repositories.

Food detection Algorithm-2022

Trained FAIR’s detectron2 for instance segmentation using MASK R-CNN which produced third position with 38.1 mAp in aicrowd’s Food recognition benchmark 2022 challenge. Finished third and fifth, respectively, in rounds 1 and 2 of this competition.

InterviewPlatform-2022

A few UI features were created as a part of the Infosys interview platform UI which completed 1 million interviews by October of 2022. And created and developed Marketplace (recruiting platform) for both recruiters and candidates.

Speech To Text-2021

Finetuned the Deepspeech model for Indian English's automatic speech-to-text translation for Infosys Recruitment platform.