Sadi Mohammad Siddiquee

I'm a PhD aspirant with a background in computer vision, specializing in medical imaging and high-resolution image analysis.

I've worked as full-time Research Assistant in CCDS lab, IUB and mHealth lab, BUET, Bangladesh. Already published couple of my works. Also contributing to our non-profitable initiative Bengali.AI to address challenges related to Bengali language through open-source dataset and research. Excited to connect and explore new opportunities in tech and research!

Email  /  CV  /  Bio  /  Google Scholar  /  Twitter  /  Github  /  Linktree

My research transects the domains of computer vision, machine learning, medical imaging, optimization and image processing. I have also worked with Ultra-high resolution image ex. Satellite Image and Contrastive learning. My representative projects are listed below.

A benchmark datset for multi-target classification of handwritten Bengali Graphemes, with novel implications for all alpha-syllabary languages, e.g., Hindi, Gujrati, and Thai.

This paper proposes an anatomy aware (AA) deep learning model that learns the generic features from x-ray images considering the underlying anatomical information. Utilizing a pre-trained model and lung segmentation masks, the model generates a feature vector including disease level features and lung involvement scores. Result: The proposed method improves the geographical extent score of Covid-19 Pneumonia Severity Prediction Dataset by 11% in terms of mean squared error (MSE) while preserving the benchmark result in lung opacity score.

This research enhances Breast Ultrasound Image Segmentation accuracy. Traditional contour models, first introduced by Kass et al., minimize an energy function using external and internal forces but rely heavily on image gradients and initialization. We address this by first using a CNN-based model (ensembled baseline) for initial localization and segmentation. While CNNs excel in localization, they lack spatial resolution and shape detail. Therefore, we feed the CNN output into a morphological snake model to achieve refined contour segmentation, eliminating initialization issues and enhancing boundary precision. This architecture yields sharper predictions, particularly for thin, small objects, and retrieves higher spatial resolution than baselines. Our method achieves state-of-the-art results on the BUSI and BUSIS benchmarks, improving mask quality (mIoU) by 6% and 11% over strong baselines.

Bengali.AI is a non-profit in Bangladesh where we create novel datasets to accelerate Bengali Language Technologies (e.g., OCR, ASR) and open-source them through machine learning competitions (e.g., Grapheme 2020, ASR 2022)

A benchmark datset for multi-target classification of handwritten Bengali Graphemes, with novel implications for all alpha-syllabary languages, e.g., Hindi, Gujrati, and Thai.

We have crowdsourced the first public 500 hr Bengali Speech Dataset on the Mozilla Common Voice platform, with speech contributed by over 20K people from Bangladesh and India.

The first large scale Multi-Domain Bengali Handwritten Digit Recognition Dataset

Yet another steal of Jon Barron's amazing website.