Solve Aero2 captchas automatically using the magic of machine learning and computer vision.
This project is intended to be run as a docker container. Prebuilt images are available on Docker Hub.
- Docker (or any other OCI compatible container runtime)
- The container needs to be able to resolve and connect to http://bdi.free.aero2.net.pl:8080/.
This project doesn't need any GPU acceleration. Even on low end hardware the solving speed is fairly quick because the captchas are small and infrequent.
you can test it by running:
docker run -it --rm dumbaspl/aero2solvershow all available options:
docker run -it --rm dumbaspl/aero2solver --helprun as a daemon that starts on boot:
docker run -d --restart=always --name aero2solver dumbaspl/aero2solverIt is also possible to run this container on RouterOS using the container package. This removes the need for a separate machine to run the solver on.
The training data was collected by marking up 500 training + 100 validation captchas by hand.
The model was trained for 20000 iterations starting from the yolov4-tiny pre-trained weights.
This project is licensed under the MIT License - see the LICENSE file for details.
- Aero2 for providing "free" internet access 😉
- darknet-rust for providing a Rust wrapper for darknet
SlavesFriends for marking up all the training data