Portfolio

Worked as a Research Associate at Computational Imaging Lab, IIT Madras, supervised by Prof. Kaushik Mitra, and closely collaborated with Prof. Aswin Sankaranarayanan.
Developed a Cross-attention network for low-light image enhancement and restoration for Autonomous systems.
Used Thermal images as a guide for RGB images to improve the performance of the enhancement algorithm.
Presented our own collected real-world V-TIEE dataset consisting of 40 RGB and thermal image pairs.
Significant improvements over the previous networks using only RGB images were observed.

Worked on the transformation of videos with the help of Generative AI technologies.
Developed a system to detect and extract the landmarks from the face of a particular individual in a video.
Conversion and enhancement of voice in audios were performed using the GAN-based FreeVC model.
Incorporated DINet model in the system for performing visual face dubbing on the high-resolution videos.
The performance of the final model improved by 17% compared to the earlier one and later deployed on the system.

Worked as a Mitacs Globalink Research Intern at Dalhousie University, Halifax where Prof. Andrew Rutenberg supervised me.
Worked on a model for classifying which health variables are of low dimensions and of high dimensions for death.
Used variational autoencoder and RNN for the model and used the English Longitudinal Study of Ageing (ELSA).
Optimized the loss function for every health variable for the model.

Worked as an Applied Research Fellow at the Centre of Visual Information Technology Lab, IIIT Hyderabad, where I was guided by Prof. Ravi Kiran Sarvadevabhatla and Prof. C V Javahar.
Developed a model to detect vehicle lights and then classify the status of the light on the Indian Driving Dataset.
Tested IDD on Facebook Detic model and then used YoloV3, YoloV4, and YoloV5 for detecting vehicle lights.
Used color thresholding on the RGB images for detecting the status of vehicle lights.
Acheived an mAP of 88.4% for the detection of vehicle lights and also performed well for classifying the status of lights.

Worked as a Research Intern at the Centre of Artificial Intelligence and Robotics Lab, DRDO.
Worked on a project to detect pedestrians and then predict their trajectories from the JAAD and PIE datasets.
Used Pedestron for the detection of pedestrians for both first-person view (FPV) and bird’s-eye view (BEV) scenarios.
Used conditional variational autoencoder (CVAE) with recurrent neural networks (RNNs) to encode observed trajectories and decode future trajectories.
The detection and prediction results were good and the whole system was deployed on a Robot.