Tahir Hasan

Tahir Hasan

Computer Vision & ML researcher — MSc CVRML, University of Surrey

I build vision systems that make fine-grained decisions efficiently. My current research uses hierarchical, coarse-to-fine models for gastrointestinal endoscopy — a router network that sends each image to the right specialist classifier, improving diagnostic accuracy by over 12% versus a flat baseline. Before Surrey, I built a 9,010-video continuous Bangla Sign Language dataset (530 word classes) and a real-time recognition pipeline — now under review — and published a first-author paper on explainable plant-disease identification at IEEE ICCIT 2025. I'm averaging 90.5% across the MSc so far.

Email CV (PDF) GitHub LinkedIn Scholar ORCID

See the BdSL recognition demo  ·  Explore the dataset

Placeholder monogram — photo coming soon

Selected publications

All publications →
Diagram of a two-stage pipeline: a YOLO detector localises the leaf, an ECA-NFNet classifier names the disease, and an explainability map highlights the evidence

An Explainable AI based Plant Disease Identification using a Two-Stage Detection-Classification Pipeline with YOLO and ECA-NFNet Framework

T. Hasan, et al.

IEEE ICCIT 2025 International Conference on Computer and Information Technology

Two-stage pipeline: YOLO localises the leaf, ECA-NFNet classifies the disease, and saliency overlays show which regions drove each prediction.

DOI Code Write-up

Hand-landmark skeleton over a video frame, feeding a temporal model that outputs a running word sequence

Implementation of Bangla Sign Language (BdSL) Recognition in Continuous Real Time

T. Hasan, M. Ahmed, S. I. Moyeen

Under review Under review

Continuous, real-time BdSL recognition built on a new 9,010-video dataset spanning 530 word classes.

Dataset Write-up

Selected projects

All projects →