TinyWOD dataset targets high-altitude open-water rescue detection
Researchers introduce TinyWOD, an annotated dataset of lifebuoys photographed from above 1.5 km, built to train and test detectors where rescue objects shrink to a few pixels.
In plain words
A group publishing on arXiv released TinyWOD, a labelled collection of aerial images for open-water search and rescue. The images are taken from high altitude, so the lifebuoys used as targets appear as tiny objects that are hard for detection software to find. The dataset is meant to help researchers build and compare models for this setting.
What was announced
Aerial platforms flying above 1.5 km can cover water more efficiently during search and rescue, but the authors of the arXiv abstract note that long imaging distances, wide fields of view and limited sensor resolution reduce targets to a few pixels, making them more vulnerable to environmental interference than conventional small objects. They state that existing datasets focus on low-altitude scenarios and, as far as they know, do not support automatic detection at 1.5-2.0 km. TinyWOD addresses this gap using lifebuoys as detection targets, with 11,359 manually annotated images and 19,505 annotated instances spanning varied environments, lighting, water colours and surface ripples.
The dataset's average equivalent side length is 12.7 pixels, and 76.96% of instances fall below 15 pixels, which the authors say reflects real high-altitude water-rescue imaging. They also define APtiny, a detection metric for tiny objects, and benchmark mainstream and state-of-the-art networks with it alongside conventional metrics. In their evaluation, all tested networks detected the tiny waterborne objects, which they present as evidence of the dataset's reusability and utility. The abstract positions TinyWOD as a resource for research on tiny object detection in high-altitude open-water search and rescue.
What it does not show
- The evidence is the paper's abstract only; it reports the authors' own annotations, metric definition and benchmark results, which have not been independently confirmed.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.