Matiur Rahman Minar

I am currently a PhD student and Research Assistant at Sogang University, focusing on computer vision and GenAI research, advised by Prof. Unsang Park. I completed MS from SeoulTech, where I worked on computer vision and deep learning under the supervision of Prof. Heejune Ahn. Previously, completed my Bachelor in CSE from BUET. In the past, I have worked as a Deep Learning Engineer at Alteredverse, and AI Engineer at NetTargets.

Email  /  Google Scholar  /  GitHub

Profiles  /  LinkedIn  /  X (Twitter)  /  Social

I'm interested in computer vision and deep learning, especially in 3D Vision and Generative AI. Below are the highlighted publications.

We propose a dual-memory framework that balances stability and motion to sustain high background persistence and continuous fluid dynamics over multi-minute horizons for fixed-camera nature video generation.

We proposed a novel approach for image-based multi-pose virtual try-on (VTON) for fashion clothing, utilizing a hybrid method for 3D clothing reconstruction from single image.

We proposed a novel hybrid and fully automatic method for 3D clothing reconstruction from single image and applying it to image-based virtual try-on (VTON) for fashion clothing, which generates realistically deformed try-on results with superior quality.

We proposed a novel hybrid and fully automatic method for 3D clothing reconstruction from single image and applying it to image-based virtual try-on (VTON) for fashion clothing, which generates realistically deformed try-on results with the highest possible quality.

We proposed a novel hybrid method for 3D clothing reconstruction and applying it to image-based virtual try-on (VTON) for fashion clothing, which generates realistically deformed try-on results.

We proposed CP-VTON+, a new pipeline for fully image-based virtual try-on (VTON) for fashion clothing, solving the limitations of state-of-the-art VTON approaches.

These include awards, challenges, and competitions.

Worked on design and implementation of a multi-pose guided image-based virtual try-on framework, trained on MPV dataset. We were evaluated 1st on SSIM metric, and 2nd on AMT study.

These include coursework, side projects, and regionally published, and unpublished research works.

Master's thesis on the fashion-clothing based online virtual try-on project, exploring a hybrid approach to preserve the realism in the try-on output.

We proposed a novel hybrid method for 3D clothing reconstruction and applying it to image-based virtual try-on (VTON) for fashion clothing, which generates realistically deformed try-on results.

We proposed a new pipeline for fully image-based virtual try-on for fashion clothing.

We analyzed and compared performances of the state-of-the-art image-based virtual try-on methods, their strengths and weaknesses.

We reconstructed 3D fashion model out of single image and generated fashion-show walk motion-capture video.

Worked on design and implementation of a robust pipeline for detecting new Korean vehicle number plates.

Semantic segmentation using state-of-the-art and improved models for fashion clothing datasets e.g. ATR, CFPD, LIP.

An overview of Deep Learning, the state-of-the-art, recently proposed models and frameworks, and applications.

Template styles merged from Jon Barron, Leonid Keselman, Antoine Guédon, and Seunghun Oh.