Anis Sarker Email & Phone Number

Dhaka, Bangladesh

• Researched and analyzed state-of-the-art algorithms and models for a virtual try-on system, focusing on Generative Adversarial Networks (GANs). Conducted extensive literature review to identify relevant pre-trained models and methodologies.• Gathered relevant data for clothing generation from diverse clothing brands worldwide to prepare a comprehensive dataset for desired models. Utilized Python along with Scrapy and Selenium to scrape and extract data from various clothing brand websites.• Developed a PyQt5 application to automate the alignment of naming conventions between models and clothing items for cloth try-on processes, eliminating manual tasks and enhancing workflow efficiency.• Implemented codes for background removal, human body pose estimation, and extraction of body joints/keypoints coordinates utilizing machine learning algorithms including rembg, segment-anything, densepose, graphonomy, and openpose to prepare tailored datasets.• Implemented GAN model within a Machine Learning Operations (MLOps) framework. Utilized intricate strategies such as integrating pixel-wise loss and posture awareness loss functions to finely warp cloth images in alignment with human pose estimations. These advanced techniques were implemented using PyTorch and other Python frameworks/libraries, ensuring robust model training, evaluation, and testing processes.• Deployed the model onto the server using Flask, enabling API services.

Dhaka, Bangladesh

• Conducted in-depth analysis on advanced deep learning models, focusing on transfer learning and optimized architectures. Prepared and pre-processed satellite and aerial imagery for training, evaluating and testing models.• Undertook in-field GPS data collection and processing to ensure the accuracy and reliability of collected geo-spatial data for research purposes on slum areas for building categorization project. Cleaned, geo-referenced, processed, and analyzed geo-spatial data, ensuring data integrity and reliability.• Created a dataset of whole Dhaka city, downloaded, and processed satellite imagery. Oversaw and guided a team of 24-25 students in satellite image segmentation, classification utilizing eCognition software.• Performed detailed land use and land cover segmentation analysis for Bangladesh, revealing insights into land utilization patterns and environmental changes in diverse category.• Conducted collaborative research with the University of Tokyo and the United Nations University on urban categorization methods, utilizing deep learning models such as DeeplabV3+ and geospatial analysis techniques through Python, PyTorch, and GIS software like QGIS and ArcGIS. Analyzed factors such as population density, land use patterns, and infrastructure development to classify urban areas into distinct categories, offering valuable insights for urban planning and development strategies.• Contributed to a project analyzing various disaster images, including landslides, fires, earthquakes, floods, droughts, avalanches, explosives, and bombing attacks, and applied machine learning algorithms to identify salient features, mimicking human attention patterns, to support disaster response efforts.• Configured and managed GPU-enabled High-Performance Computing (HPC) and Super Computing clusters, optimizing computational resources for efficient data processing and analysis.

Dhaka, Bangladesh

• Responsible for sourcing spatial data from satellite imagery providers such as Sentinel-2 and Landsat-8 to support geo-spatial analysis projects.• Curated benchmark datasets, including Gaofen-2 and ISPRS Potsdam, for training machine learning models. Conducted research on GIS studies focusing on classification, and segmentation in categories like built-up, forest, water body, farmland etc.• Implemented FCN-8 and VGG-19 models to analyze satellite imagery, achieving a significant improvement in accuracy with a result of 84\%. This achievement played a pivotal role in our first publication, demonstrating the practical application of our research findings. Prepared and submitted manuscripts detailing research findings for potential publication in conferences and journals.