Python
I love using real-life examples and offering guidance in understanding the core concepts in computational methods. I work on many projects and papers, contributing and taking responsibility as a co-owner of the project. My research career in augmented reality, image segmentation, regression, and classification shows my skills. I'm interested in AI research because I believe it can change education and solve real-world problems. My goal is to solve research problems in these areas and make a positive impact in the academic and research community. I also try to contribute to the community by providing weekly lectures to needy students at a local non-profit charity organization.
Operating System
Machine Learning
Research expertise
1. Flight Fare Prediction
We developed a flight fare prediction model using a comprehensive Kaggle dataset containing 300,153 rows and 12 columns, capturing various factors influencing airline ticket prices such as flight duration, days until departure, and departure/arrival times. We implemented several regression models, including Linear Regression, Decision Tree Regressor, Bagging Regressor, Random Forest Regressor, Support Vector Regressor (SVR), K-Neighbors Regressor, Extra Trees Regressor, Ridge Regression, and Lasso Regression.
2. Basic Image Segmentation
We developed an image segmentation model using a state-of-the-art deep learning library to accurately segment images. The project aimed to classify each pixel in an image into predefined categories, enhancing the ability to interpret visual data. We utilized advanced segmentation models from the `segmentation_models` library, including UNet, PSPNet, LinkNet, and FPN.
3. Vision Transformer for Remote Sensing Image Classification
We explored Vision Transformers, which use multihead attention mechanisms instead of convolution layers, to derive long-range contextual relations between pixels in images. Our model achieved high classification accuracy on Merced, AID, and Optimal31 datasets, outperforming state-of-the-art approaches.
4. nnU-Net for Medical Image Segmentation
We used nnU-Net on datasets from the Medical Segmentation Decathlon challenge, Automatic Cardiac Segmentation Challenge (ACDC), and LiTS to understand its versatility compared to UNet. nnU-Net demonstrated robust performance across different medical imaging tasks.
5. Segment Anything for Medical Segmentation
We evaluated the Segment Anything Model (SAM), trained on over 1 billion annotations for natural images, on several medical imaging datasets using point and box prompts. Our results show that SAM performs well on well-circumscribed objects but struggles with complex tasks like brain tumor segmentation. Box prompts yielded better performance than point prompts, demonstrating SAM's impressive zero-shot segmentation capabilities for some medical datasets but moderate to poor performance for others.
Python
Tensorflow
Keras
Linux
Microsoft Office
Dr. Milton Mondal
Institute of Neurophysiology
University Medical Center Göttingen
milton.mondal@med.uni-goettingen.de
Prof. Dr. Shreemoyee Ganguly
Professor, University of Engineering and Management
shreemoyee.ganguly@uem.edu.in
https://scholar.google.com/citations?user=ayOHO2AAAAAJ&hl=en&oi=ao