Structured Cross-Scale Attention for Hierarchical Vision Transformer Backbones
Manuscript under review
This work introduces structured cross-scale attention as an efficient feature-fusion module for RT-DETR-style object detectors. It targets a common limitation of hierarchical vision transformer backbones: preserving useful information across scales without adding unnecessary architectural complexity. Evaluation across COCO, VisDrone, and multiple backbone settings studies the method's effect on accuracy, efficiency, and robustness.
Object DetectionDETRVision TransformersMulti-Scale FeaturesEfficient Detection
Explaining Transformer-Based Road Damage Detection
Manuscript under review
This work examines road damage detection as a structurally difficult object detection problem: defects are often small, elongated, visually ambiguous, and inconsistently annotated. It compares CNN- and DETR-family detectors on RDD-2022 and uses explainability analysis to study why transformer-based detectors may better fit the visual and spatial demands of road damage detection.
Object DetectionVision TransformersDETRExplainabilityRoad Damage Detection
Spatial feature fusion in 3D convolutional autoencoders for lung tumor segmentation from 3D CT images
Suhail Najeeb, Mohammed Imamul Hassan Bhuiyan
Biomedical Signal Processing and Control, 78
A lung tumor segmentation study introducing spatial feature fusion in 3D convolutional autoencoders to improve volumetric CT segmentation performance.
Medical ImagingSegmentationDeep Learning
A Pipeline for Lung Tumor Detection and Segmentation from CT Scans Using Dilated Convolutional Neural Networks
Shahruk Hossain, Suhail Najeeb, Asif Shahriyar, Zaowad R. Abdullah, M. Ariful Haque
ICASSP 2019 - IEEE International Conference on Acoustics, Speech and Signal Processing
An automated pipeline for lung tumor detection and segmentation in 3D CT scans using dilated convolutional neural networks and post-processing.
Medical ImagingSegmentationCNNs
Classification of Retinal Diseases from OCT scans using Convolutional Neural Networks
Suhail Najeeb, Nowshin Sharmile, Ipsita Sahin, Mohammad Tariqul Islam, Mohammed Imamul Hassan Bhuiyan
2018 10th International Conference on Electrical and Computer Engineering
A retinal OCT classification study using region-of-interest detection and convolutional neural networks to identify retinal abnormalities from scans.
Medical ImagingOCTCNNs