Arnab Ghosh | Rochester Institute of Technology

About Me

I am currently pursuing a Master’s in Imaging Science from Rochester Institute of Technology, New York. My anticipated graduation date is 2023.

My current research focus is on Shift Variant Image Deconvolution. I have developed a novel deep learning-based image restoration algorithm rooted in physics-informed principles which is called ShiVaNet. It has proven highly effective in nullifying shift-variant image blur, surpassing contemporary techniques with a notable improvement of 0.2 dB in PSNR. My research is relevant to microscopic images and can unblur highly blurred images given the point spread function of the microscope.

Previously,I worked as a Senior Engineer in the Computer Vision team at L&T Technology Services for 3 years. I played a pivotal role in spearheading the development of Bird Eye Object Detection. This involved leading the integration and calibration of cameras for generating a bird’s eye view image and developing a highly efficient deep learning pipeline for almost real-time object detection on this unique perspective.

Additionally, I made substantial contributions to diverse projects, including Activity Recognition on Jetson TX2 board, Visual Odometry, and Piping and Instrumentation diagram digitization.

I have a Bachelors of Engineering degree in Instrumentation and Electronics Engineering from the reputed Jadavpur University in India.

Research Interests

Computer Vision: Image Restoration, Object Detection , Super-Resolution
Machine Learning: Physics Informed Machine Learning, Self Supervised Learning,Normalizing Flow, Vision Transformers

Skills

Programming Languages: Python, C/C++, MATLAB
Deep Learning Frameworks: PyTorch, PyTorch Lightning, TensorFlow
Deep Learning: CNN, RNN, LSTM, Vision Transformers, Generative Adversarial Networks, Auto-Encoders
Computer Vision: Image Processing, Image Restoration, Image Segmentation
Machine Learning & Numerical Computing Libraries: OpenCV,NumPy, Scikit-learn,Keras
Mathematics and Statistics: Linear Algebra, Calculus, Finite Element Methods, Probability and Statistics
Tools& Version Control: Weights and Biases, Jupyter, VS-Code, Spyder, Git, GitHub
Parallel Programming: CUDA
Computational Imaging: Fourier Optical Simulation, Computer Generated Hologram

News

[Nov. 2023] My paper was awarded the Best Student Paper Award at WNYISPW
[Aug. 2021] Joined Rochester Institute of Technology as PhD Student.
[Mar. 2020] Awarded “Employee of the Month” for my contribution to the development of Bird Eye View Object Analytics.
[Aug. 2018] Joined Larsen and Toubro Technology Services as a Senior Engineer

Publications

IEEE WNYISPW

ShiVaNet:Shift Variant Deconvolution using Deep Learning

Arnab Ghosh*, Grover Swartzalnder

IEEE/Western New York Image And Signal Processing Workshop (WNYISPW), 2023.

Code Early Accept

IEEE ASPCON

Prefilter-Equalizer Based Structure: A New Design Strategy for Narrow-band FIR Filters

Arnab Ghosh, Subhabrata Roy , Abhijit Chandra

IEEE Applied Signal Processing Conference ASPCON, 2018.

Project Page Conference Paper