Research
Publications
Published, accepted, and preprint work across computer graphics, medical imaging AI, and formal verification.
07 entries
- Venue
- Conference PaperPreprintJournal Article
- Areas
- Speech and Language ProcessingComputer GraphicsFormal VerificationMedical Imaging AI
- 01
A Glossary Must Be Scoped to the Speech It Will Hear: Terminology Control for Live Thai-English Medical Captioning
A deployed real-time Thai-English captioning system for medical conferences built around a deterministic term-protection layer: one curated glossary grounds an LLM correction stage over ASR finals, and protected terms are masked before a black-box MT call and restored afterwards, with an append-on-loss fallback. On 440 synthesized utterances with known ground truth, the best configuration reaches 90.0% term fidelity against 47.9% for corrected text through plain MT; on 54.1 minutes of real speech, a support check blocks 19 of 20 off-domain glossary insertions.
BibTeX
@inproceedings{sriwaranon2026glossary, title = {{A Glossary Must Be Scoped to the Speech It Will Hear: Terminology Control for Live Thai-English Medical Captioning}}, author = {Teejuta Sriwaranon and Monchai Lertsutthiwong}, year = {2026}, booktitle = {2026 21st International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)}, note = {Accepted, to appear} } - 02
Cross-Specimen Forecasting of Copper Oxidation Appearance for Physically Based Rendering
A system that observes a copper plate with a fixed camera, forecasts its appearance ten accelerated units ahead, and converts the forecast into metallic-roughness PBR maps. Under leave-one-recording-out evaluation across an accelerated-chamber and an open-air recording, a training-free global affine color extrapolation is the only method that beats repeating the last observed frame in both directions (+11.7% and +51.2%), while chamber-trained learned models improve only on held-back frames of their own recording.
BibTeX
@inproceedings{sriwaranon2026crossspecimen, title = {{Cross-Specimen Forecasting of Copper Oxidation Appearance for Physically Based Rendering}}, author = {Teejuta Sriwaranon and Borworntat Dendumrongkul and Tanapat Chamted and Sirichai Leelachao and Pizzanu Kanongchaiyos}, year = {2026}, booktitle = {2026 30th International Computer Science and Engineering Conference (ICSEC)}, note = {Accepted, to appear} } - 03
Timed-Automata Verification of Payload-Link Supervision, Watchdog Gating, and A/B Boot Policy in STM32 Flight Software
Verifies an abstracted design model of three interacting recovery mechanisms in an STM32 small-satellite flight-software prototype (payload-link supervision, a health-gated hardware watchdog, and a CRC-validated A/B bootloader) as UPPAAL timed automata. Across fifteen model configurations and 254 TCTL formulas every verdict matches a pre-declared oracle; model checking exposes one design defect, confirmed and fixed in source, and the verified bounds are checked against hash-verified hardware traces from 90 fault-injection trials.
BibTeX
@inproceedings{dendumrongkul2026timedautomata, title = {{Timed-Automata Verification of Payload-Link Supervision, Watchdog Gating, and A/B Boot Policy in STM32 Flight Software}}, author = {Borworntat Dendumrongkul and Natdanai Hirunsirikul and Teejuta Sriwaranon and Phongsatorn Saisutjarit and Pizzanu Kanongchaiyos}, year = {2026}, booktitle = {2026 30th International Computer Science and Engineering Conference (ICSEC)}, note = {Accepted, to appear} } - 04
Formal Verification of DirectX11 Graphics Pipeline Using Colored Petri Nets
A compositional Colored Petri Net model of all ten stages of the DirectX 11 graphics pipeline, with the core rasterization path exhaustively verified against six LTL properties spanning safety, liveness, boundedness, and resource correctness. Breadth-first exploration reaches up to 208,905 states; a fault-injection matrix shows every checker is sensitive to its own fault, and two properties fail in stronger forms: per-lane starvation freedom under an arbitration policy DX11 permits, and token conservation under geometry-shader fan-out.
BibTeX
@inproceedings{dendumrongkul2026formal, title = {{Formal Verification of DirectX11 Graphics Pipeline Using Colored Petri Nets}}, author = {Borworntat Dendumrongkul and Teejuta Sriwaranon and Nuengwong Tuaycharoen and Wiwat Vatanawood}, year = {2026}, booktitle = {2026 30th International Computer Science and Engineering Conference (ICSEC)}, note = {Accepted, to appear} } - 05
What Will This Copper Look Like Later? Forecasting Surface Appearance and Rendering It as a PBR Material
A system that forecasts how copper surfaces evolve as they oxidize and renders the predicted appearance as a PBR material (albedo, normal, roughness, metallic) from a single camera observation. A learned spatio-temporal model underperforms last-frame copying on unseen specimens, while a closed-form global color extrapolation transfers successfully, improving accuracy by 13.4-16.7% depending on the prediction horizon, indicating that learned models encode specimen-specific corrosion patterns whereas global color trajectories generalize.
BibTeX
@misc{sriwaranon2026copper, title = {{What Will This Copper Look Like Later? Forecasting Surface Appearance and Rendering It as a PBR Material}}, author = {Teejuta Sriwaranon and Borworntat Dendumrongkul and Tanapat Chamted and Pizzanu Kanongchaiyos}, year = {2026}, eprint = {2608.28102}, archivePrefix = {arXiv}, primaryClass = {cs.GR}, doi = {10.48550/arXiv.2608.28102}, url = {https://arxiv.org/abs/2608.28102} } - 06
An End-to-End Deep Learning Pipeline for Automated Mandible Virtual Surgical Planning Using Real-World Clinical Data
An end-to-end deep learning pipeline that automates mandible virtual surgical planning from real-world clinical data, combining volumetric segmentation and reconstruction to streamline the workflow for maxillofacial surgery.
BibTeX
@article{kamboonsri2026endtoend, title = {{An End-to-End Deep Learning Pipeline for Automated Mandible Virtual Surgical Planning Using Real-World Clinical Data}}, author = {Nattapon Kamboonsri and Teejuta Sriwaranon and Natdanai Tantisereepatana and Chedtha Puncreobutr and Boonrat Lohwongwatana and Gregory B. Olson and Alessandro Tel and Massimo Robiony and Titipat Achakulvisut and Peerapon Vateekul}, year = {2026}, journal = {IEEE Access}, doi = {10.1109/ACCESS.2026.3702327}, url = {https://ieeexplore.ieee.org/document/11557298} } - 07
Fine-Grained Formal Verification of an Asynchronous Speaker Diarization Pipeline Using Hierarchical Timed Colored Petri Nets
A fine-grained hierarchical Timed Colored Petri Net (HTCPN) model of an asynchronous speaker diarization pipeline that exposes three generalizable structural deadlock patterns in concurrent AI pipelines. Seven correctness properties are verified, including a novel attribution consistency property, showing that sub-module decomposition is necessary for complete pipeline verification.
BibTeX
@inproceedings{sriwaranon2026finegrained, title = {{Fine-Grained Formal Verification of an Asynchronous Speaker Diarization Pipeline Using Hierarchical Timed Colored Petri Nets}}, author = {Teejuta Sriwaranon and Nuengwong Tuaycharoen and Wiwat Vatanawood}, year = {2026}, booktitle = {2026 23rd International Joint Conference on Computer Science and Software Engineering (JCSSE)}, doi = {10.1109/JCSSE68839.2026.11597080}, url = {https://ieeexplore.ieee.org/document/11597080} }