Research Projects

Timeline: Spring'26 - present

I am developing an O-HIC framework in this project with a special focus on effectively utilizing auxiliary information (e.g., attribute data) during classification task.

Timeline: Spring'25 - Fall'25

I developed an HIC framework in this project with a special focus on generalizability.  I conducted thorough experiments with various HIC benchmark datasets and also utilized attribute information in this HIC task. Currently, this is an under review article in ECCV-26 conference.

Timeline: Fall'25

I developed a InfoNCE-style contrastive learning loss that can incorporate physics-based and general shape-based cues (e.g., Histogram-oriented gradients (HOG)), and helps extracting usable object boundaries in foreground-background consistency problem in IR images. Currently, this is an under review article in ICPR-26 conference.

Timeline: Summer'25 - Fall'25

Funding: U.S. Department of Agriculture–Agricultural Research Service (USDA–ARS)

I collaborated with my friend and lab-mate Zirak Khan,  where he developed HyperRoPE-SST, a transformer-based architecture based on a patch-local mixed-axis 2-D RoPE with learnable frequencies that preserves geometric relationships amongst the pixels within a hyperspectral patch while adapting to dataset-specific spatial spectral characteristics. We published a research article in the IEEE Journal Of Selected Topics In Applied Earth Observations and Remote Sensing. 

Timeline: Summer'25 - Fall'25

Funding: National Science Foundation (NSF)

I developed a hierarchical image benchmark dataset named Coral-Net, and subsequently used this dataset in my HIC projects. This project was a sub-part from the 3D Coral Image Reconstruction project.

Timeline: Summer'23 - Fall'24

Funding: Department of Defense (DoD) 

I worked in this project to develop a robust image classifier for the Army. Specifically, the goal was to build a classifier that was trained solely with clean images while still being a reliable performer in the battlefield faced with stressed images (e.g., distortion, camouflage, etc.).  We also aimed to integrate various auxiliary information which could positively help us for this stressed battlefield conditions. 

Initially, I collaborated with friend and lab-mate Sandipani Basu, on this project and published a research article in the Colour and Visual Computing Symposium (CVCS). We developed a model called CoNNText, where we fuse shape-based cues (shape context) to guide during classification in the stressed imaging conditions. In the later phase, I continued in this project and developed another model called ViTAtr, and it was published in International Symposium on Visual Computing (ISVC). In this project, I worked on integrating attribute-based auxiliary information with visual data and effectively improved classification performance in stressed imaging conditions.