Aditi Prabakaran

Aditi Prabakaran

Undergraduate Computer Vision & AI Researcher · SRM Institute of Science and Technology

My research currently focuses on optimizing 3D Gaussian Splatting to enable faster rendering of 3D scene reconstructions through:

  • Transient Removal: Segmenting dynamic occlusions via CLIP-guided supervision to recover clean background geometry.
  • Primitive Processing: Applying per-Gaussian scoring and pruning to eliminate ghosting artifacts and reduce memory footprint.

My long-term research goal is 3D computer vision for Embodied AI, developing 3D world models that enable autonomous systems and robots to perceive and interact with dynamic physical environments.

ap0232@srmist.edu.in Chennai, India
GitHub LinkedIn CV

Research Experience

International Institute of Information Technology, Hyderabad Jun 2025 – Present
Research Assistant (3D Vision and Reconstruction)

Designed and implemented a semantic-guided 3D Gaussian Splatting (CLIP-GS) framework leveraging vision-language supervision for transient object detection and removal.

Indian Institute of Technology, Kharagpur Mar 2026 – Aug 2026
Research Intern (Multi-Agent Reinforcement Learning)

Developed a cooperative multi-agent reinforcement learning framework using QMIX for adaptive quick-commerce route optimization under dynamic traffic, weather, and operational constraints.

Indian Academy of Sciences, Bengaluru Jul 2025 – Aug 2025
Summer Research Fellow (3D Vision & Neural Rendering)

Benchmarked multi-model rendering pipelines (NeRF, InstantNGP, 3DGS) on consumer hardware (RTX 3050, 6GB VRAM) to systematically evaluate trade-offs in VRAM overhead, memory footprint, and rasterization speed under resource constraints.

Publications

Semantic-Guided 3D Gaussian Splatting for Transient Object Removal
arXiv preprint (arXiv:2602.15516) (2026)
Aditi Prabakaran, Priyesh Shukla
Targeted Data Augmentation for Hallucination Mitigation and Faithfulness Enhancement in Compact LLMs
13th International Conference on Computing for Sustainable Global Development (INDIACom) (2026)
Aditi Prabakaran, Thatavarthy Sai Anvitha, Harine G.K
Scalable Remote Risk Estimation for Bridge Infrastructure: A Latent Variable Probabilistic Approach
3rd International Conference on Advances in Artificial Intelligence and Machine Learning in Big Data, Springer (2026)
Aditi Prabakaran, Shreyansh Jain, Joy Saha

Achievements

2026
Selected among top 40 contributors nationwide for Code for GovTech (C4GT) DMP 2026.
2026
Won 5th Place among 400+ teams at StudAI One Foundry Hackathon.
2025
Selected as a Top 5 Finalist out of 50,000+ applicants in Smart India Hackathon (SIH 2025).
2025
Awarded the prestigious IAS Summer Research Fellowship (selection rate under 2%).