=== Important dates ===
Method Submission Deadline: 6th June, 2025
Contest Paper Deadline: 13th June, 2025
========================

=== Contest ===
Throughout history, societies have faced fire-related risks, which
intensified during the industrial era due to machinery malfunctions and
misuse. Today, fire remains a major threat to human life, infrastructure
and ecosystems. To prevent disasters and protect the environment,
authorities are turning to advanced surveillance systems powered by
Computer Vision algorithms for automatic, reliable fire detection. Early
Computer Vision approaches, based on color and motion models, struggled
with the variability of real-world scenes. The introduction of Machine
Learning and Deep Learning techniques significantly improved detection
performance, though challenges persist due to the complex nature of fire
phenomena and limitations in available datasets. Detection failures
often occur when fires appear differently from the training samples, for
example when visible from greater distances or when moving objects
resembling fire confuse the system, leading to false alarms. A review of
the literature highlights two main gaps in current methods. The first
concerns the need to design detection systems according to the
application scenarios. While well-trained, frame-based detectors perform
effectively in simple situations where flames or smoke are clearly
visible and no other moving objects are present, more complex scenarios
— such as when flames are small or numerous moving objects resemble fire
— require sophisticated models incorporating temporal analysis
techniques. Enhancing methods with scenario awareness and tailoring them
to specific operational conditions can significantly improve real-world
performance. The second gap relates to achieving an optimal balance
between precision and recall. Although current methods show good
sensitivity in detecting fires (high recall), they often lack precision
in distinguishing fire from visually similar objects. This issue was
also evident during the first ONFIRE 2023 contest, where even
top-performing systems generated excessive false alarms, undermining
operational reliability and increasing costs due to the need for human
intervention. In this context, the ONFIRE 2025 international competition
has been launched to foster the development of advanced, real-time fire
detection algorithms for fixed CCTV cameras, deployable on smart cameras
or embedded systems with limited resources. The contest challenges
participants to create solutions that address these limitations across
four application scenarios of varying difficulty:

– Low Activity – Short Range (easy)
– Low Activity – Long Range (intermediate)
– High Activity – Short Range (difficult)
– High Activity – Long Range (intermediate)

Each method will be evaluated on a private test set of unseen,
scenario-categorized videos and ranked both overall and by scenario.
Additionally, frame processing speed and memory usage will be assessed
to ensure efficiency and resource compatibility. A final score,
combining F1-score with resource consumption, will determine the
official rankings. Competitors will work with an expanded dataset
compared to ONFIRE 2023, featuring over 300 annotated videos from public
sources, with the option to incorporate additional publicly available
data. A reference baseline will also be provided for performance
comparison.

The detailed description can be read here:
https://mivia.unisa.it/onfire2025/
========================

=== Rules ===
The deadline for the submission of the methods is 6th June, 2025. The
submission must be done with an email in which the participants share
(directly or with external links) the trained model, the code and the
report. The participants can receive the training set and its
annotations by sending an email, in which they also communicate the name
of the team. The participants can use these training samples and
annotations but also additional videos. The participants are strongly
encouraged to submit a contest paper by the deadline of 13th June, 2025.
The paper can be submitted through Easychair. The maximum number of
pages is 12 including references. Accepted papers will be included in
the ICIAP 2025 Workshops Proceedings.

The detailed instructions can be read here:
https://mivia.unisa.it/onfire2025/
========================

The organizers,

Diego Gragnaniello, University of Salerno, Italy
Antonio Greco, University of Salerno, Italy
Carlo Sansone, University of Naples – Federico II, Italy
Bruno Vento, University of Naples – Federico II, Italy