Chaitali Bhattacharyya

Education

PhD in Electrical Engineering and Computer Science

DGIST, Daegu, South Korea, 2024 – Present

Master of Electronics Engineering

Yeungnam University, Daegu, South Korea, 2022 – 2024

Korean Language

Kangwon National University, Chuncheon, South Korea, 2021 – 2022

Bachelor of Electronics and Instrumentation Engineering

NiT, WBUT, Kolkata, India, 2017 – 2021

Skills

Technical: Python, C++; PyTorch, TensorFlow, OpenCV, LLMops, Matplotlib, Seaborn, Plotly; Linux, CUDA

Research Areas: Efficient AI, CV, LLM, Multi-modal Learning, Generative AI

Research Experience

Graduate Student Researcher

Advanced Visual Intelligence Lab, South Korea, Sept 2022 – Aug 2024

• Developed architectures with diffusion models for point cloud generation and for detection and classification, increasing accuracy by 5-7%.
• Developed hybrid CNN model to detect black ice through Hyperspectral Imaging with 97% accuracy.
• Collaborated with multiple research organizations and companies on robot vision projects for industrial purposes. [News]
• Optimised YOLO models for jetson-nano to detect industrial boxes collaboration with PHA,South Korea

Research Intern

UP Lab, CVSSP, Surrey University, UK, Dec 2023 – April 2024

• Curated efficient and robust deepfake dataset from huge diffusion-based deepfake datasets.
• Proposed robust training strategies for more generalized models.
• Increased accuracy by 20% compared with other SOTA models. [ArXiv]

Research Intern

Defense Research Lab (DRDO), India, Dec 2020 – April 2021

• Applied ML techniques to enhance water purification analysis.

Projects (Research Institute, Industry Collaboration)

• SLM: Developing efficient and small LLM based on Sparse Autoencoder’s importance score.
• Point Cloud Generation: Collaborated with Yeungnam University Hospital, ETRI, DGIST on real-time 3D pose estimation and point cloud processing.
• Pose Estimation: Developed 3D/2D human pose estimation method using point clouds, with ETRI and DGIST.
• Cooperative Robot: Created a detection model for aiding robots in logistics, in collaboration with PHA.
• ARUCO Marker Detection: Implemented ARUCO marker detection and overlay for VARJO XR3 in a mixed reality setup.

Papers

• Diffusion Deepfake, CVPR-W
• Black Ice Classification with Hyperspectral Imaging, Applied Sciences, MDPI
• EGCY-Net: An ELAN and GhostConv-Based YOLO Network for Stacked Packages in Logistic Systems, Applied Sciences, MDPI
• S-LIGHT: Synthetic Dataset for Reflection Separation, Sensors, MDPI

Patent

Artificial Neural Network for Automated Box Recognition, Box Recognition Method and Computing Device, and Recording Medium Thereof, Reference No: 2024000047

Awards and Honors

• DGIST’24 Fellow: Fully funded Government Scholarship for PhD
• Global Korea Scholarship (GKS’21): Funded scholarship for higher study (MS-Electronics Engineering, South Korea)