Ashraful Islam
Ashraful Islam

Computer Vision & Deep Learning Engineer

NVIDIA

I am a computer vision and deep learning engineer at NVIDIA. I 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.

Download CV

Experience

Computer Vision & Deep Learning Engineer

NVIDIA

  • Led end-to-end development of production 2D/3D perception models, from data curation through deployment with DLA integration and in-vehicle validation.
  • Architected multi-task 2D detection pipeline, implementing pruning and quantization-aware training for optimized on-device inference.
  • Spearheaded 3D lane and object detection pipeline using self-supervised training with SAM-based auto-labeling, reducing annotation costs.
  • Designed novel KPIs for targetless camera-lidar calibration evaluation, enabling quality assessment without ground truth (ICRA 2023).

Graduate Research Assistant

Rensselaer Polytechnic Institute

Developed frameworks for cross-domain few-shot learning, weakly supervised temporal action localization, and deep learning models with hybrid attention mechanism (HAMNet). Built systems for automatic tracking and anomaly detection at airport security checkpoints.

Research Intern

Microsoft Research

Worked on self-supervised multi-task representation learning for 3D computer vision, focusing on camera pose estimation and segmentation tasks.

Research Intern

IBM Research

Researched contrastive representation learning for transfer learning. Proposed joint objective of self-supervised contrastive loss with cross-entropy for better transferability.

Research Intern

Kitware Inc.

Developed DOA-GAN for image and video copy-move forgery detection and localization.

Education

PhD in Computer Engineering

Rensselaer Polytechnic Institute (RPI)

Thesis supervised by Prof. Richard Radke. Research focused on cross-domain few-shot learning, weakly supervised action localization, and deep metric learning.

MS in Computer Engineering

Rensselaer Polytechnic Institute (RPI)

BS in Electrical Engineering

Bangladesh University of Engineering and Technology (BUET)

Skills
Technical Skills
Python
PyTorch
Deep Learning
Computer Vision
C++
Research Areas
Few-shot Learning
Self-supervised Learning
Action Detection
Object Tracking
Publications
A Islam, R Chen, R Panda, L Karlinksky, RS Feris, RJ Radke (2021). Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data. NeurIPS.
Code PDF Poster Slides
A Islam, R Chen, R Panda, L Karlinksky, RJ Radke, RS Feris (2021). A Broad Study on the Transferability of Visual Representations with Contrastive Learning. ICCV.
Code PDF Video
A Islam, R Chen, R Panda, L Karlinksky, RJ Radke, RS Feris (2021). A Simple Framework for Cross-Domain Few-Shot Recognition with Unlabeled Data. CVPRW.
PDF PDF Poster Slides Video
A Islam, C Long, RJ Radke (2021). A Hybrid Attention Mechanism for Weakly-Supervised Temporal Action Localization . AAAI.
Code PDF Poster
A Islam, C Long, A Basharat, A Hoogs (2020). DOA-GAN: Dual-Order Attentive Generative Adversarial Network for Image Copy-move Forgery Detection and Localization. CVPR.
PDF Poster Video
A Islam, RJ Radke (2020). Weakly Supervised Temporal Action Localization Using Deep Metric Learning. WACV.
Code PDF Poster Video
A Islam, Y Zhang, D Yin, O Camps, RJ Radke (2018). Correlating Belongings with Passengers in a Simulated Airport Security Checkpoint. ICDSC.
PDF Video
Awards
Best Paper Award
ICDSC 2018 ∙ September 2018
Awarded for research on computer vision and deep learning applications.