Ashraful Islam

Ashraful Islam

Graduate Research Assistant

Rensselaer Polytechnic Institute

Ashraful Islam

Hello! I am a computer vision and deep learning engineer at Nvidia. I have recently completed my PhD from the ECSE department of Rensselaer Polytechnic Institute (RPI), Troy, NY, supervised by Prof. Richard Radke. Broadly, I am interested in computer vision and deep learning. I develop neural network models for unsupervised, semi-supervised, weakly-supervised, and few-shot learning. I have also worked on action detection and object tracking.

Previously, I spent three wonderful summers at Microsoft Research, IBM Research and Kitware. I completed my undergrad in Electrical Engineering from BUET, Bangladesh.

My Resumé.

Interests
  • Computer Vision
  • Deep Learning
  • Self-supervised Learning
Education
  • PhD in Computer Engineering, 2021(Expected)

    Rensselaer Polytechnic Institute (RPI)

  • MS in Computer Engineering, 2019

    Rensselaer Polytechnic Institute (RPI)

  • BS in EE, 2017

    Bangladesh University of Engineering and Technology (BUET)

Experience

 
 
 
 
 
Research Intern
May 2021 – Aug 2021 WA
My work focuses broadly on self­-supervised multi­task representation learning for 3D computer vision. Specifically, I am working on self-supervised pretraining of deep neural networks that provide better representations for downstream camera pose estimation and segmentation task.
 
 
 
 
 
Research Intern
Jun 2020 – Aug 2020 NY
I researched on unsupervised and supervised contrastive representation learning for transfer learning. Our study suggests that networks trained with contrastive loss is more transferable to a different domain than the networks trained with supervised cross-entropy loss. Proposed and analyzed joint objective of self-supervised contrastive loss with cross-entropy or supervised contrastive loss that leads to better transferability of these models over their standard-trained counterparts.
 
 
 
 
 
Research Intern
May 2019 – Aug 2019 NY
I developed a deep adversarial model titled ‘Dual-order Attentive Generative Adversarial Network (DOA-GAN)’ for image and video copy-move forgery detection and localization.
 
 
 
 
 
Graduate Research Assistant
Aug 2017 – Present NY
  • Developed a framework for cross-domain few-shot learning that uses unlabeled images from novel dataset during meta-training.
  • Proposed a weakly supervised temporal action localization method using metric learning that only requires video-level action instances as supervision during training. Also developed a deep learning model with hybrid attention mechanism (HAMNet) for solving the issues of action completeness and background modeling in temporal action localization with weak supervision, outperforming SOTA methods by atleast 2.8%.
  • Developed a system that automatically tracks passengers and items, and detects unusual activities (baggage theft, left-behind items, etc.) at an airport security checkpoint (demo video).

Publications

Recent News

  • Mar 2022: Passed Doctoral Defense Exam.
  • Sep 2021: One paper accepted to NeurIPS 2021.
  • July 2021: Our paper on contrastive learning accepted to ICCV 2021.
  • May 2021: Started summer internship at Microsoft Research.
  • April 2020: Passed Doctoral Candidacy Exam.
  • April 2020: Paper accepted as oral to CVPR L2ID workshop.
  • Dec 2020: Paper accepted to AAAI 2021.
  • June 2020: Started summer internship at IBM Research.
  • Feb 2020: Paper accepted to CVPR 2020.
  • Dec 2019: Paper accepted to WACV 2020.
  • May 2019: Started summer internship at Kitware Inc.
  • Sep 2018: Best paper award in ICDSC 2018.

Contact