Datasets:
Request access to Z-FPV-v1
Z-FPV-v1 contains annotated battlefield footage of grounded "waiter" (ambush) FPV drones. Access is reviewed manually by the authors and is granted for non-commercial academic research only.
By requesting access you agree that: (1) you will use the data solely for non-commercial academic research; (2) you will not redistribute the data or any derivative that allows reconstruction of the source frames; (3) you will not use the data to develop or improve offensive or targeting systems; (4) you will cite the accompanying paper in any resulting publication.
Log in or Sign Up to review the conditions and access this dataset content.
Z-FPV-v1
Z-FPV-v1 is an initial dataset for vision-based detection of grounded "waiter" (ambush) FPV drones (aka "zhdun") — first-person-view drones that are flown to a location, landed, and left in low-power standby to wait for a passing vehicle.
Unlike existing counter-UAV benchmarks, which target airborne drones seen against the sky, Z-FPV-v1 captures the inverted geometry: a motionless, camouflaged target on the ground, imaged from a ground / UGV viewpoint, frequently at night or in low light.
Representative sequences
Figure 1: Representative Z-FPV-v1 sequences. Thermal low-light, small-object waiter-FPV, and daylight conditions are shown from top to bottom. Frames t₁–t₅ are sampled at one-second intervals. Best viewed zoomed in.
Supported tasks
- Object detection — 2D bounding-box detection of a single class (
waiter_fpv_drone). - Secondary use cases: small-object detection, camouflaged object detection, low-light / thermal detection, domain-shift and robustness studies under downlink compression artifacts.
Data collection
Because "waiter" FPV drones are a very recent phenomenon and no public dataset of them exists, the collection was assembled from open-source intelligence: publicly available footage reposted on social media in which such drones appear.
Annotation
- All footage was annotated in CVAT using tight 2D bounding boxes.
- A semi-automatic pipeline was used: a few key frames were labeled by hand, the transformer tracker TransT propagated boxes across intermediate frames, and the propagated labels were then manually corrected.
- For quality control, authors independently reviewed every annotation, with disagreements resolved by discussion — yielding a doubly reviewed label set.
Intended use and ethical considerations
Z-FPV-v1 is released to support protective and defensive research: earlier automatic detection of ambush drones threatening logistics convoys, medical evacuation, command vehicles, personnel rotations, and increasingly civilian transport. Once a "waiter" is noticed, crews often have only minutes — sometimes seconds — to react.
Permitted: non-commercial academic research on detection, robustness, and sensing for threat avoidance.
Not permitted: targeting, strike, or other offensive applications; redistribution of the data or derivatives that reconstruct the source frames; any use that violates the terms of the original sources.
No personally identifying information is intentionally included. If you identify PII in the release, contact the authors and it will be removed.
Access and licensing
This is a gated dataset. Access requests are reviewed manually by the authors; approval is not automatic.
Citation
TBA
- Downloads last month
- 26