Welcome to the Unidrone v1.0 AI models 🚁🎉
What is Unidrone?
Unidrone is a a set of detection AI models, based on a large YOLO-v8 backbone and my own data pipeline. It's mostly a merging of two of my other models: WALDO and NANO. The former is great at detections in nadir imagery (straight overhead) and the latter is better for "forward-looking" angles. Before you had to choose one of those models and stick to a limited set of angles, but now Unidrone does both really well!
The AI models are capable of detecting these classes of items in drone imagery ranging in altitude from about 50 feet to 1000 feet, regardless of angle:
For a video demonstration, see:
0 -> 'Person' --> people! all kinds of people including ones that are on bikes or swimming in the sea 🧍♀️🕺💃🧜🏽♀️🏂🧞
1 -> 'Bike' --> bikes, mopeds, motorbikes, all stuff with 2 wheels 🚲
2 -> 'LightVehicle' --> all kinds of civilian cars, including pickup trucks, vans etc... 🚗🏎️🚓🚐🚑
3 -> 'Truck' --> large commercial vehicles including articulated trucks or big box-on-chassis delivery trucks 🚚
4 -> 'Bus' --> a bus 🚌
5 -> 'Boat' --> boats, ships, canoes, kayaks, surf boards... all the floaty stuff 🚢🏄
6 -> 'Digger' --> all kinds of construction vehicles, including tractors and construction gear 🚜
For users of previous versions of my models: note that I removed the military class and smoke detection. This is meant to be a FOSS tool for civilian use and I don't want to pursue making it work for military applications. Safety applications however like border monitoring are very much supported!
What is it good for?
People are currently using versions of Unidrone for:
disaster recovery
monitoring wildlife sanctuaries (intruder detection)
occupancy calculation (parking lots etc..)
monitoring infrastructure
SAR
crowd counting
ground risk mitigation (avoiding things on the ground)
Let me know what you build!
Of course depending on your deployment platform you may want to export the models in various different formats such as ONNX or TFlite. The models are published here in .pt format so they can be exported very easily, for instance using the Ultralytics Yolov8 implementation here.
AI model details
It's a set of YOLOv8 model, trained on my own datasets of synthetic and augmented / semi-synthetic (overlayed layers of synthetic stuff on normal images) data.
I'm not going to release the dataset for the time being.
The weights are completely open in .pt format, allowing you to deploy in any number of ways!
How can I start using Unidrone?
Check out the boilerplate code in the repo to run the models and output pretty detections on your drone videos using the wonderful Supervision annotation library from Roboflow :)
Going deeper
Of course if you know your way around deploying AI models there is a lot more you do
with this release, including:
fine-tuning the models on your own data
building a nicely optimized sliding-window inference setup that works super-duper-fast on your own edge hardware
quantizing the models for ultra performance on cheap devices
using the models to annotate your own data and train something of your own
Enjoy!
SUPPORT
Note that these AI models are meant for experts who know how to use them. Before you purchase them you should know how to run a Yolov8-format detection AI model on the hardware you want to use. While I will answer the occasional question to help people deploy things I am not offering any long-term support with these models, and don't have the bandwidth to help people deploy in every possible environment.
DETECTION OF SOMETHING ISN'T WORKING AS EXPECTED
If you'd like to support further development of the model by sending data of things that don't quite work yet please feel free to contact me at:
LICENSE TLDR
These models are under MIT license. This means you can use them for whatever you want, including commercial usage. It also means you can distribute them to others, if you feel so inclined. I would appreciate if you purchased them if you find any commercial value in using these AI models however, than way I can continue making them better :)
Unless otherwise specified all code in this release is published with the
licence conditions below.
MIT License
Copyright (c) 2025 Stephan Sturges / Aircortex.com
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.