Kunal
Bhosikar

I am a final-year dual-degree B.Tech + M.S. by Research student in Computer Science and Engineering at IIIT Hyderabad, working under the guidance of Dr. Charu Sharma in the Machine Learning Lab.

My research lies at the intersection of 3D computer vision, computer graphics, and digital humans. I am broadly interested in modeling physically plausible human behavior and interaction in 3D environments, including human-scene interaction, grasp-aware motion synthesis, co-speech gesture generation, scene-aware human motion planning, hand-hair interaction, and digital human simulation. I also work on geometry processing, 3D reconstruction, neural rendering, and the robustness of generative 3D/4D pipelines. My work focuses on applications in AR/VR, embodied AI, and virtual humans.

I am actively looking for PhD positions in 3D computer vision, graphics, and digital humans.

Digital Humans Human–Scene Interaction Fine-grained Human Interaction Motion & Gesture Generation Scene-Aware Motion Planning Neural Rendering Geometry Processing Robustness of 3D/4D Generative Models

News

  • May 2026 Successfully defended my MS Research thesis — Towards Realistic Scene-Aware Human-Object Interaction in 3D World — at IIIT Hyderabad, advised by Dr. Charu Sharma 🎉.
  • Apr 2026 Two papers accepted at CVPR 2026 Workshops: Fast and Robust Mesh Simplification for Generated and Real-World 3D Assets 🏆 Best Paper Runner-up at the 3D4S Workshop, and PatchPoison: Poisoning Multi-View Datasets to Degrade 3D Reconstruction at the SPAR-3D Workshop.
  • Jan 2026 Took a hands-on session on Multi-Layer Perceptron (MLP) architectures for NHAI (National Highway Authority of India) employees as a part of the Division of Flexible Learning (DFL) initiative by IIIT Hyderabad.
  • Nov 2025 Paper accepted at WACV 2026 ★ OralInteracTalker: Prompt-Based Human-Object Interaction with Co-Speech Gesture Generation.
  • Sep 2025 Paper accepted at BMVC SU2HM Workshop 2025 ★ OralMOGRAS: Human Motion with Grasping in 3D Scenes.
  • Aug 2025 Completed a research internship at TCS Research, Delhi — developed a fast quadric-based mesh simplification framework for large-scale VR scenes, working under the guidance of Dr. Lokender Tiwari.
  • Aug 2025 Started as a Teaching Assistant for the Software Systems Development course at IIIT Hyderabad, Fall 2025 semester.
  • Jul 2025 Indian Patent granted — System and Method for Detecting Exercise Poses and Counting Repetitions (with Dr. Charu Sharma et al.).

Publications

* Equal contribution  ·  Bold = Kunal Bhosikar

CVPR Workshop · 2026 ★ Oral 🏆 Best Paper Runner-up

Fast and Robust Mesh Simplification for Generated and Real-World 3D Assets

Bhosikar, K., Savalia, P., Tiwari, L., Bhowmick, B.

The 2nd Edition of 3D Geometry Generation for Scientific Computing (3D4S) @ CVPR 2026

CVPR Workshop · 2026

PatchPoison: Poisoning Multi-View Datasets to Degrade 3D Reconstruction

Wadekar, P., Bhosikar, K., Bachina, V.S.P., Gangwal, A., Sharma, C.

Security, Privacy, and Adversarial Robustness in 3D Generative Vision Models (SPAR-3D) @ CVPR 2026

WACV · 2026 ★ Oral

InteracTalker: Prompt-Based Human-Object Interaction with Co-Speech Gesture Generation

Rajan, S.*, Bhosikar, K.*, Sharma, C.

IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2026

BMVC Workshop · 2025 ★ Oral

MOGRAS: Human Motion with Grasping in 3D Scenes

Bhosikar, K., Katageri, S., Madhavaram, V., Han, K., Sharma, C.

1st Workshop on From Scene Understanding to Human Modeling (SU2HM) @ BMVC 2025

Indian Patent · 2025

System and Method for Detecting Exercise Poses and Counting Repetitions

Sharma, C., Bhosikar, K., et al.

Research Projects

Selected ongoing and completed projects. Published work listed above.

ENACT

Under Review · NeurIPS 2026

w/ Nikos Athanasiou  ·  Max Planck Institute

A training-free framework for synthesising scene-aware human motion from a single RGB image and language instructions. Chains foundation models for monocular scene reconstruction, affordance grounding, pose optimisation, and motion diffusion to produce contact-consistent, physically plausible interactions.

Foundation Models Scene Interaction Motion Synthesis

Single-View 3D Hair Reconstruction

Ongoing

w/ Vanessa Sklyarova  ·  MPI  ·  ETH Zürich  ·  TU Darmstadt

Building upon Im2Haircut to reconstruct strand-based 3D hair geometry from single images and monocular video. Extending the pipeline for SMPL-X bodies with realistic hand–hair interaction, geometric consistency, and physically plausible contact modelling.

3D Hair SMPL-X Digital Humans

Watermarking in 4D Gaussian Splatting

Ongoing

w/ Nicole Meng  ·  Tufts University

Investigating hidden-scene embedding and watermarking techniques for dynamic 4DGS representations. Exploring robust and imperceptible encoding strategies using Gaussian and spherical harmonics attributes, inspired by GS-Hider.

4DGS Watermarking Neural Rendering

Dense 3D Reconstruction via Stereo

Implemented Semi-Global Matching and PatchMatch for dense depth estimation from stereo image pairs. Evaluated on the KITTI Stereo dataset, analysing accuracy–efficiency trade-offs for large-baseline outdoor configurations.

Stereo Vision Depth Estimation KITTI

Imitation Learning for UAV Obstacle Avoidance

Implemented behaviour cloning and DAgger using expert velocity-obstacle-based policies for UAV navigation. Analysed generalisation and failure modes in cluttered indoor environments.

Imitation Learning Robotics DAgger

Dynamic Graph CL for Medical Report Generation

Developed a framework integrating dynamic knowledge graphs and contrastive learning for automated radiology report generation. Achieved state-of-the-art results on the IU-Xray dataset.

Medical AI Contrastive Learning Knowledge Graphs

Experience

Undergraduate Researcher

Machine Learning Lab  ·  IIIT Hyderabad

May 2023 – Present
  • Research in 3D Computer Vision & Graphics under Dr. Charu Sharma, with international collaborators from MPI, ETH Zürich, Tufts University, HKU, and TU Darmstadt.
  • Focus areas: human-scene interaction, grasp-aware motion synthesis, diffusion-based motion generation, 3D human hand-hair interaction, and adversarial attacks on 3DGS/4DGS.
  • Involved in dataset creation, benchmarking, and evaluation protocol design for scene-aware human motion generation.

Research Intern

TCS Research  ·  Delhi, India

May – Aug 2025
  • Worked under Dr. Lokender Tiwari. Developed a fast quadric-based mesh simplification framework for large-scale VR scenes with geometry and appearance preservation.
  • Designed curvature-aware boundary handling and texture-preserving error metrics — leading to a publication at CVPR 3D4S Workshop 2026 (Oral).

Teaching & Service

Teaching Assistant

3D Vision Summer School 2024

IIIT Hyderabad & IIIT Bangalore

Teaching Assistant

Software Systems Development

IIIT Hyderabad  ·  Fall 2025

Mentor

Data Structures & Linear Algebra

IIIT Hyderabad