Skip to content
Sabid Mahmud, home
Contact

Sabid Mahmud · B.Sc. in Computer Science, BRAC University

Research

I study interpretable computer vision for medical imaging, with a focus on attention-based deep learning for cervical cytology.

Research interests

  • Computer vision
  • Medical image analysis
  • Interpretable machine learning
  • Weakly supervised learning
  • Dataset curation & evaluation
Email me Request academic CV

PapSmear Dataset

  1. PapSmear Dataset

    A dataset of 2,253 expert-annotated Pap-smear images from 300+ source slides, for computational cytology research.

    My contribution: Co-curated the dataset with the research team and developed the workflow for independent specialist annotation.

Research projects

Undergraduate thesis · BRAC UniversityJanuary 2025 – June 2026

Interpretable Attention-Based Multiple Instance Learning for Cervical Cancer Detection

Studied how attention-based deep learning can classify high-resolution Pap-smear images while retaining local cellular detail and showing where the model focused.

My contribution

  • Contributed to the team’s deep learning pipeline using full-resolution image patches, a ResNet-50 encoder, and gated-attention multiple instance learning.
  • Built PapLens so people can run inference with our cervical-cytology deep learning models and inspect the resulting attention maps in the original field.

Thesis grade: A (94.3%).

Computer vision study · BRAC University2026

Wildlife Species Classification

Compared convolutional networks and a vision transformer on the severely imbalanced iWildCam 2022 camera-trap dataset.

My contribution

  • Evaluated ResNet-50, EfficientNet-B0, and ViT-B/16 using sequence-aware splits to prevent temporal and burst-image leakage.
  • Examined the trade-off between overall classification performance and recognition of rare species.

Research prototype. Not for clinical or diagnostic use.

Research software

PapLens

An application for running the cervical-cytology deep learning models from our research and inspecting the resulting attention maps.

VeraLabel

Self-hosted image annotation and independent review software, developed from the thesis annotation workflow.