|
Hi! I am a final-year master's student at Texas A&M, currently working as a Graduate Research Assistant in the Visual and Spatial AI Lab under Prof. Suryansh Kumar. Previously, I worked as a backend developer at Deloitte for two years before starting my master's. Prior to that, I earned my Bachelor's degree in Electrical and Electronics from Manipal Institute of Technology, advised by Prof. Harish Kumar J.R.. During my final year, I was also selected in Google Summer of Code (GSoC). My research interests lie at the intersection of Machine Learning, LLM's/VLM's, and computer vision. I am particularly interested in dense prediction problems such as optical flow, stereo and depth estimation, as well as perception tasks involving VLM's and LLM's. My current work involves models including Vision Transformers (ViT-16 and Swin), DINOv2, and PaliGemma-2. Email / Github / Linkedin / CV I am actively seeking both full-time and summer-internship opportunities. |
|
|
|
|
|
Praroop Chanda, Suryansh Kumar Developing a unified model for optical flow, stereo, and depth estimation using continual learning to enhance cross-task knowledge transfer and efficiency. |
|
|
Praroop Chanda, Rutwik Palaskar, Suryansh Kumar Extending Freestyle Layout-to-Image Synthesis (FLIS) to 3D by integrating spatial masks and multi-view consistency cues for generating coherent 3D structures or multi-view 2D renderings. |
|
|
|
Praroop Chanda, J.H. Gagan, B Vaibhav Mallya, Harish Kumar J.R. IEEE INDICON, 2023 | Link Automated optic cup segmentation method using affine snakes in a gradient vector field. |
|
|
|
Praroop Chanda, Ishaan Rawal Report / GitHub Optimized contrastive learning under small batch constraints using HyperOpt with ASHA scheduler. SogCLR with AdamW outperformed CLIP and CyCLIP, achieving 28.45% Mean Recall on COCO and 55.99% Top@10 accuracy on ImageNet. |
|
Praroop Chanda, Rutwik Palaskar Meghana Chintalapati Report / GitHub Developed a multimodal predictive framework for calorie estimation by integrating meal images, CGM data, and Viome features. Utilized ResNet50, BERT, and cyclic encoding, training with a custom RMSRE loss function to achieve a 0.2294 score. |
|
Supervised by Puneet Sharma and Pradeeban Kathiravelu Google Summer of Code (GSoC), 2022 GitHub Developed two key modules for MRI Quality Control analysis: High Contrast Spatial Resolution and Low Contrast Object Detectability. Utilized image preprocessing, NCC, multi-Otsu thresholding, and Hough Transform techniques to extract and analyze MRI phantom features. |
|
Last updated: March 2025
|