§R — Research index
Research output, in the open.
Preprints, peer-reviewed papers, datasets, and registered software from the lab — each with its venue, its numbers, and a link to the source. Filter by type or area, or search the abstracts.
Showing 8 of 8
- [01]Peer-reviewedAcceptedIEEE · ICAITech 2026Sep 19, 2026
CLC-YOLO: A Compact Channel-Gated Prototype Network for Real-Time Leakage-Aware Breast Ultrasound Lesion Segmentation
M. Fazri Nizar, Muhammad Naufal Rachmatullah, Julian Supardi
To appear · preprint arXiv:2609.31702
- Δ Dice
- +3.11 pp · BUS-UCLM · gains on all 4 datasets
- Overhead
- +64 params · +0.0966 ms on T4 (FP16 TensorRT)
Abstract
Reliable breast ultrasound lesion segmentation requires accurate boundaries and evaluation that prevents patients or duplicate images from crossing data splits. We propose Channel Local Contrast (CLC), a compact refinement of the YOLO26 segmentation prototype head. CLC adds a fixed local high-pass residual controlled by 64 zero-initialized, bounded channel gates. Baseline and CLC were compared in five matched folds on each of four breast ultrasound datasets. BUS-BRA used patient-disjoint outer tests with separate inner validation. BUS-UCLM and BrEaST used patient-grouped validation folds; BUSI used duplicate-component groups because patient identifiers are unavailable. Group-macro Dice increased by 1.68, 3.11, 1.12, and 2.45 percentage points on BUS-BRA, BUS-UCLM, BUSI, and BrEaST, respectively. Only the BUS-BRA paired 95% confidence interval excluded zero. CLC adds 64 parameters and 0.0049 giga floating-point operations (GFLOPs). At 640 pixels, single-T4, batch-one, 16-bit floating-point (FP16) TensorRT graph times were 2.1478 ms for CLC and 2.0512 ms for baseline, excluding preprocessing and postprocessing. CLC increased group-macro Dice across all four datasets with a measured T4 forward-pass overhead of 0.0966 ms.
Computer Vision · Medical Imaging - [02]PreprintarXiv · cs.CVJun 6, 2026
Aqua Boundary-Saliency Attention Module for Lightweight Underwater Salient Instance Segmentation Detection Transformer
M. Fazri Nizar, Julian Supardi, Muhammad Naufal Rachmatullah
arXiv:2606.08002
- Latency
- 4.31–6.34 ms · NVIDIA T4 · FP16 TensorRT
- Benchmarks
- 4 · UIIS · UIIS10K · USIS10K · USIS16K
Abstract
Underwater instance segmentation integrates pixel-level mask prediction and instance-level discrimination for marine resource exploration, ecological monitoring, and underwater robotic perception. Recent prompt-based and auxiliary-modality methods improve mask quality, but their reliance on large foundation models, prompt generation, or extra modality estimation complicates efficient deployment. This work introduces Lightweight Underwater Salient Instance Segmentation Detection Transformer (LUSIS-DETR), a compact detection-transformer framework built around the Aqua Boundary-Saliency Attention Module (AquaBSAM). AquaBSAM embeds underwater boundary, contrast, attenuation, chroma, dark-channel, and center-prior cues into DINOv2-initialized multi-scale features through bounded residual modulation, while auxiliary mask supervision and small-object copy-paste are training-only. Extensive evaluation on four recent underwater instance segmentation datasets, UIIS, UIIS10K, USIS10K, and USIS16K, shows competitively leading performance against previous state-of-the-art works across category-aware and salient-instance protocols. TensorRT half-precision (FP16) benchmarking on an NVIDIA T4 graphics processing unit (GPU) achieves 4.31-6.34 milliseconds (ms) latency, supporting real-time inference under an accessible reproduction setting.
Computer Vision · Marine & Underwater - [03]DatasetHarvard DataverseApr 2026
Inter-Patient Split Fetal Head Ultrasound Segmentation Dataset in YOLO Polygon Format
M Fazri Nizar
DOI 10.7910/DVN/FKXHLL
- Images
- 3,832 · inter-patient 70/15/15 split
- Structures
- 3 · Brain · CSP · lateral ventricles
Abstract
This dataset provides an inter-patient stratified train/val/test split (70/15/15) of 3,832 fetal head ultrasound images with YOLO-format polygon segmentation labels for three anatomical structures: Brain, Cavum Septi Pellucidi (CSP), and Lateral Ventricles (LV). The source images originate from the HC18 Grand Challenge (van den Heuvel et al., 2018) and the large-scale fetal head biometry annotation dataset (Alzubaidi et al., 2023; Data in Brief, vol. 51, art. no. 109708), which provides 959×661 px B-mode frames with polygon masks verified by a senior attending physician and a radiologic technologist (ICC ≥ 0.859). Splitting is performed at the patient level, all frames from a given patient appear in exactly one partition, to prevent data leakage that inflates reported scores when frames from the same patient appear in both training and test sets. The split preserves approximate class stratification and is seeded (seed = 42) for full reproducibility. The dataset contains 7,664 files organized as follows: - train/: 2,654 images + 2,654 labels (6,551 total instances across all splits) - val/: 603 images + 603 labels - test/: 575 images + 575 labels Instance counts per split: - Train: 2,697 Brain, 921 CSP, 1,052 LV (4,670 total) - Val: 611 Brain, 190 CSP, 212 LV (1,013 total) - Test: 568 Brain, 172 CSP, 212 LV (952 total) Labels follow the YOLO polygon segmentation format: each line contains class_id followed by normalized polygon vertices (x1 y1 x2 y2 … xN yN). A dataset.yaml configuration file and split_info.json (patient-to-partition mapping) are included. This dataset accompanies the paper "Domain-Guided YOLO26 with Composite BCE-Dice-Lovász Loss for Multi-Class Fetal Head Ultrasound Segmentation" (arXiv:2603.26755).
Computer Vision · Medical Imaging · Datasets & Benchmarks - [04]PreprintarXiv · cs.CVMar 23, 2026
Domain-Guided YOLO26 with Composite BCE-Dice-Lovász Loss for Multi-Class Fetal Head Ultrasound Segmentation
M. Fazri Nizar
arXiv:2603.26755
- Mean Dice
- 0.9253 · vs 0.9012 published SoTA baseline
- Inference
- Prompt-free · detect + segment in one pass
Abstract
Segmenting fetal head structures from prenatal ultrasound remains a practical bottleneck in obstetric imaging. The current state-of-the-art baseline, proposed alongside the published dataset, adapts the Segment Anything Model with per-class Dice and Lovász losses but still depends on bounding-box prompts at test time. We build a prompt-free pipeline on top of YOLO26-Seg that jointly detects and segments three structures, Brain, Cavum Septi Pellucidi (CSP), and Lateral Ventricles (LV), in a single forward pass. Three modifications are central to our approach: (i) a composite BCE-Dice-Lovász segmentation loss with inverse-frequency class weighting, injected into the YOLO26 training loop via runtime monkey-patching; (ii) domain-guided copy-paste augmentation that transplants minority-class structures while respecting their anatomical location relative to the brain boundary; and (iii) inter-patient stratified splitting to prevent data leakage. On 575 held-out test images, the composite loss variant reaches a mean Dice coefficient of 0.9253, exceeding the baseline (0.9012) by 2.68 percentage points, despite reporting over three foreground classes only, whereas the baseline's reported mean includes the easy background class. We further ablate each component and discuss annotation-quality and class-imbalance effects on CSP and LV performance.
Computer Vision · Medical Imaging - [05]DatasetHarvard DataverseMar 2026
Indonesian Cross-Domain Benchmarks for Semantic Parsing and Text-to-SQL
M. Fazri Nizar, Abdiansah Abdiansah
DOI 10.7910/DVN/C5AO0A
- Language
- Bahasa Indonesia · cross-domain Text-to-SQL
Abstract
This dataset provides a standard benchmark for Indonesian Text-to-SQL research and cross-domain semantic parsing. Adapted from the well-established English Spider dataset, it addresses the underrepresentation of Indonesian in natural language interfaces for databases. The dataset pairs natural language questions in Bahasa Indonesia with complex SQL queries, developed through a rigorous four-stage pipeline of extraction, translation, cleaning, and conversion to ensure strict SQL syntax and database integrity.
Applied NLP · Datasets & Benchmarks - [06]Registered IPIndonesia · DJKI2025
Risk Meter Deterministic Algorithm for Website Security Assessment
M. Fazri Nizar, Harisman Nizar, Nabil Pasha
Reg. No. 001042181
Software Security - [07]Registered IPIndonesia · DJKI2025
FazScan: Cross-Platform Software Application as Scanner of Information and Security of Websites
M. Fazri Nizar, Harisman Nizar
Reg. No. 000849276
Software Security - [08]Peer-reviewedIEEE · ICIC 2024Oct 24, 2024
IDSpider: Indonesian Standard Dataset for Text-to-SQL
Abdiansah Abdiansah, Novi Yusliani, Fathoni Fathoni, Muhammad Fazri Nizar, Aulia Salsabella, Agi Agustian Davi
2024 Ninth International Conference on Informatics and Computing (ICIC)
- Benchmark
- IDSpider · first Indonesian Text-to-SQL benchmark
Abstract
This research presents IDSpider, a new standard dataset for Text-to-SQL research in Indonesian, developed by translating the well-established Spider dataset into Indonesian. Text-to-SQL is a critical technology that bridges the gap between natural language processing (NLP) and database management systems, allowing non-technical users to retrieve data from databases using simple, natural language queries. However, most existing datasets are in English, leaving nonEnglish languages, especially Indonesian, underrepresented in the field. This paper describes the challenges encountered in translating both natural language questions and SQL queries, given the need for precision in maintaining SQL syntax and database structure integrity. We implemented a four-stage process consisting of Extraction, Translation, Cleaning, and Conversion to generate the IDSpider dataset. Furthermore, we tested the performance of the BRIDGE model on both the original Spider dataset and the translated IDSpider dataset. Results show that while the BRIDGE model performs well on Spider, its performance drops significantly when applied to IDSpider, primarily due to language differences and translation complexities. This research establishes the first benchmark for Indonesian Text-to-SQL tasks and lays the groundwork for further improvements in cross-lingual natural language interfaces for databases.
Cited by 1 · Applied NLP · Datasets & Benchmarks