Anti-UAV
A benchmark and toolkit for discovering, detecting, recognizing, and tracking UAVs in the wild using RGB and thermal infrared video.
What is Anti-UAV?
Anti-UAV is a research project and benchmark suite for detecting and tracking unmanned aerial vehicles (drones) in real-world environments. It provides datasets, evaluation metrics, and baseline models to address the computer vision task of discovering, detecting, recognizing, and tracking UAV targets using RGB and/or thermal infrared video. The project aims to mitigate security risks from unauthorized drone intrusions.
Computer vision researchers and engineers working on object detection, tracking, and aerial security systems, particularly those focused on multi-modal (RGB/IR) video analysis and real-world drone surveillance applications.
It offers the first high-quality, publicly available benchmark specifically for anti-UAV tasks, with large-scale datasets capturing challenging real-world scenarios. The dual support for PyTorch and Jittor frameworks provides flexibility and optimization for different hardware environments.
Overview
🔥🔥Official Repository for Anti-UAV🔥🔥
Use Cases
Best For
- Building computer vision models for drone detection and tracking in security applications
- Researching multi-modal (RGB and thermal infrared) object tracking
- Benchmarking anti-UAV algorithms against standardized datasets and metrics
- Developing systems for airspace protection around critical infrastructure
- Studying object tracking in dynamic backgrounds with small-scale targets
- Participating in international anti-UAV challenges and workshops
Not Ideal For
- Teams needing a production-ready, out-of-the-box drone detection system for immediate deployment
- Projects focused on tracking non-UAV objects or general-purpose video surveillance
- Developers with limited GPU resources or strict constraints on model training time
Pros & Cons
Pros
Provides public datasets like Anti-UAV300 with Full HD RGB and thermal infrared videos, filling a critical gap in UAV tracking benchmarks for real-world scenarios.
Includes sequences with dynamic backgrounds, complex movements, and tiny-scale UAV targets, making it relevant for practical airspace security applications.
Offers implementations in both PyTorch and Jittor, with Jittor versions optimized for domestic hardware support and inference speed, catering to diverse user needs.
Organizes and hosts international workshops at top conferences like CVPR and ICCV, driving continuous benchmark evolution and community engagement.
Cons
The README admits problems with training in Jittor, limiting its utility for users who prefer this framework despite its hardware optimizations.
Requires specific Python versions, CUDA compatibility, and has dependency installation errors that need to be ignored, increasing setup friction and potential bugs.
The model zoo section is minimally populated with 'Keep updating...', indicating a lack of ready-to-use pre-trained models or extensive baseline implementations.
Frequently Asked Questions
Related Projects
Fast-PlannerA Robust and Efficient Trajectory Planner for Quadrotors
OpenAerialMapOpenAerialMap is an open service to provide access to a commons of openly licensed imagery and map layer services.
DroneDBFree and open source software for geospatial data storage.
Fire Detection UAVAerial Imagery dataset for fire detection: classification and segmentation (Unmanned Aerial Vehicle (UAV))
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.