Garv Kaushik [ garvkaushik1 at gmail dot com ]

Contact | GitHub | Scholar | LinkedIn | CV

I'm a pre-final Integrated Master’s student at IIT (BHU) Varanasi in Math and Computing. I love playing around with ideas in Deep Learning—especially if they're about Geometric Deep Learning or 3D Computer Vision. Right now, I’m working part-time with the awesome folks at Mixar, helping build foundational models for CG applications to make 3D design and modeling easier. In the past, I teamed up with Pinaki Nath Chowdhury from Yi-Zhe Song’s SketchX group on a fun idea around Implicit Animations for Signed Distance Fields—hoping rig-free animations become a thing. I also wrapped up a research internship with the STORM team at IRIT, Toulouse, exploring SSM-compatible neural architectures for 3D edge detection.

News & Updates

Research Papers

ViBe: A Text-to-Video Benchmark for Evaluating Hallucination in Large Multimodal Models
Vipula Rawte, Sarthak Jain, Aarush Sinha, Garv Kaushik, et al.
We introduce ViBe: a large-scale Text-to-Video Benchmark of hallucinated videos from T2V models. We identify five major types of hallucination: Vanishing Subject, Numeric Variability, Temporal Dysmorphia, Omission Error, and Physical Incongruity. Using 10 open-source T2V models, we developed the first large-scale dataset of hallucinated videos, comprising 3,782 videos annotated by humans into these five categories. ViBe offers a unique resource for evaluating the reliability of T2V models and provides a foundation for improving hallucination detection and mitigation in video generation. We establish classification as a baseline and present various ensemble classifier configurations, with the TimeSFormer + CNN combination yielding the best performance, achieving 0.345 accuracy and 0.342 F1 score. This benchmark aims to drive the development of robust T2V models that produce videos more accurately aligned with input prompts.
Proc. NAACL Workshop - TrustNLP 2025

Exploring Aapative Structure Learning for Heterophilic Graphs
Garv Kaushik
Graph Convolutional Networks (GCNs) gained traction for graph representation learning, with recent attention on improving performance on heterophilic graphs for various real-world applications. The localized feature aggregation in a typical message-passing paradigm hinders the capturing of long-range dependencies between non-local nodes of the same class. The inherent connectivity structure in heterophilic graphs often conflicts with information sharing between distant nodes of same class. We propose structure learning to rewire edges in shallow GCNs itself to avoid performance degradation in downstream discriminative tasks due to oversmoothing. Parameterizing the adjacency matrix to learn connections between non-local nodes and extend the hop span of shallow GCNs facilitates the capturing of long-range dependencies. However, our method is not generalizable across heterophilic graphs and performs inconsistently on node classification task contingent to the graph structure.
arXiv-cs.LG

Experience

• [Aug'25 - Present] MLE Intern @ Mixar
  Building the ML layer for Mixar which is an early stage start-up for AI copilot for 3D design and modelling.
• [May'25 - Aug'25] Research Intern @ STORM Team , Institut de Recherche en Informatique de Toulouse
  Supervisor: Prof. Loic barthe , Dr. Nicolas Mellado
• [July'24 - Feb'25] Research Intern @ SketchX Lab, University of Surrey
  Supervisor: Dr. Pinaki Nath Chowdhury (then PhD student of Dr. Yi Zhe Song)
• [May'24 - July'24] Summer Research Intern @ ML-Lab, Indian Statistical Institute
  Supervisor: Dr. Swagatam Das
• [Dec'23- March'24] AI-Research Intern @ mirrAR
  Supervisor: Dr. Mustafa Sameer

Communities

• Core Team Member (2023-24) @ Club of Programmers (IIT BHU) - Intelligence Group
• Core Team Member (2023-24) @ Club of Programmers (IIT BHU) - Infosec Group

Other-projects

AI-FHE Invisible Image Watermarking

VQA for GIFs

Neural-ODE

VQA for Biomed-images

Image Denoising-Autoencoders

Readings

• [NLP & Multimodal Learning]
  • Attention is All You Need (Paper )
  • Semantically Grounded QFormer for Efficient Vision Language Understanding (Paper )
• [2D/3D Computer Vision]
  • NeRF: Neural Radiance Fields (Paper )
  • MeshSDF(Paper )
  • DeepSDF(Paper )
  • Instant-NGP (Paper )
  • PCEDNet : A Lightweight Neural Network for Fast and Interactive Edge Detection in 3D Point Clouds (Paper )
  • IM Avatar: Implicit Morphable Head Avatars from Videos (Paper )
  • InstructScene : Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (Paper )
  • (VQ-VAE) Neural Discrete Representation Learning (Paper )
  • TEMOS: Generating diverse human motions from textual descriptions(Paper )
  • Towards Implicit Correspondence in SDF Evolution(Paper )
  • PCEDNet : A Lightweight Neural Network for Fast and Interactive Edge Detection in 3D Point Clouds(Paper )
  • Persistence Analysis of Multi-scale Planar Structure Graph in Point Clouds (Paper )
• GNNs & Graph Transformers
  • Semi-Supervised Classification with Graph Convolutional Networks (Paper )
  • A Generalization of Transformer Networks to Graphs (Paper )
  • NodeFormer: A Scalable Graph Structure Learning Transformer for Node Classification (Paper )