Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Towards efficient context-aware classification with compact VLM architectures: indoor fire case study.

Anh Tuan Giang1, Nhat Quang Doan2, Minh Duong Tran1

  • 1University of Science and Technology of Hanoi, Vietnam Academy of Science and Technology, 18 Hoang Quoc Viet, Nghia Do, Hanoi, Vietnam.

Scientific Reports
|April 14, 2026
PubMed
Summary

Related Concept Videos

Force Classification01:22

Force Classification

2.7K
Forces play a crucial role in the study of physics and engineering. They are essential in describing the motion, behavior, and equilibrium of objects in the physical world. Forces can be classified based on their origin, type, and direction of action.
Contact and non-contact forces are two of the most widely used categories of forces. As the name suggests, contact forces require physical contact between two objects to act upon each other. Examples of contact forces include frictional,...
2.7K
Classification of Systems-I01:26

Classification of Systems-I

700
Linearity is a system property characterized by a direct input-output relationship, combining homogeneity and additivity.
Homogeneity dictates that if an input x(t) is multiplied by a constant c, the output y(t) is multiplied by the same constant. Mathematically, this is expressed as:
700

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same journal

Peripheral B-cell receptor repertoire predicts immune-related adverse events following immune checkpoint inhibitor therapy in advanced renal cell carcinoma.

Scientific reports·2026
Same journal

Effects of black soldier fly (Hermetia illucens L.) larvae zoocompost on the mineral element content of blue honeysuckle berries.

Scientific reports·2026
Same journal

Investigation on absorption refrigeration performance of R1243zf with imidazolium ionic liquid as the working pairs.

Scientific reports·2026
Same journal

DeepTriage-CN: integrating clinical text with vital signs for emergency department admission prediction in an aging population.

Scientific reports·2026
Same journal

Gold nanoparticles as dual-action antiviral agents: disruption of SARS-CoV-2 viral envelopes and RNA integrity.

Scientific reports·2026
Same journal

Comparison of capillary microsampling and venous blood for multi-pathogen serosurveillance.

Scientific reports·2026
See all related articles
This summary is machine-generated.

This study introduces a new fire detection system using vision-language models for context-aware classification. It accurately distinguishes between dangerous and controlled fires, reducing false alarms in indoor environments.

Area of Science:

  • Computer Vision
  • Artificial Intelligence
  • Safety Engineering

Background:

  • Current deep learning fire detection systems often provide binary classification, leading to false alarms by failing to differentiate hazardous from benign fires.
  • Accurate indoor fire detection is crucial for emergency response and safety in residential and industrial settings.

Purpose of the Study:

  • To develop a lightweight and efficient framework for context-aware fire classification using vision-language models.
  • To enable nuanced differentiation between visually similar but contextually distinct fire events, moving beyond simple binary classification.

Main Methods:

  • Utilized a visual encoder to process input images, followed by a Vision Language Model (VLM) to generate natural-language captions or semantic embeddings.
  • Employed a language model to perform high-level semantic classification into 'no fire,' 'controlled fire,' and 'dangerous fire' categories.
Keywords:
ClassificationContext-aware fire classificationLoRA LLMVLM

Related Experiment Videos

  • Evaluated the framework on re-labeled public datasets and a custom ConFire dataset.

    • Main Results:

    • The proposed framework achieves high accuracy in fire classification.
    • Demonstrated significant reduction in computational overhead compared to existing methods.
    • Successfully enabled nuanced differentiation between different fire scenarios based on visual context and scene semantics.

    Conclusions:

    • Integrating vision-language reasoning into fire classification tasks is effective for enhancing safety monitoring systems.
    • The developed approach offers an intelligent and resource-efficient solution for next-generation fire detection.
    • The system's ability to classify fires into three categories significantly reduces disruptive false alarms.