Repository holds code and data for video classification model - of cars representing particular teams in Formula 1.
-
/images/: exemplary screenshots of the solution are stored here - to ilustrate how it looks and works, -
/input/: data for checking model accuracy is stored here, -
/models/: binary files with trained ML models used across the solution, -
/requirements/: folder with apt/python/binary package requirements necessary for API to work properly, -
/static/: folder for serving model-annotated files. Cleaned-up after sending any request to/predictAPI endpoint, -
/templates/: HTML templates to render withJinjaPython templating framework, -
/vision/: folder with Pytorch-related functions, -
.dockerignore: file with objects to be ignored by Docker during build, -
.gitignore: file with objects to be ignored by Git, -
constants.pyandfunctions.py: Python modules for separating code frommain.pyAPI file, -
main.py: Python file with API implementation and endpoints created withFastAPIframework, -
formula-one-video-classification.ipynb: improved / fixed jupyter notebook explaining model usage process (by tarik-yilmaz-kanzileri on Roboflow), -
README.md:Markdown-based file you are currently reading, -
environment.yaml: file with full package requirements, necessary for above-mentionedjupyter notebookto work without errors.
docker build --no-cache \
--build-arg MINIO_URL="<change_me>" \
--build-arg MINIO_ACCESS_KEY="<change_me>" \
--build-arg MINIO_SECRET_KEY="<change_me>" \
-t f1-recognition-model:v0.7 -f Dockerfile .
docker run -it \
-e MINIO_URL="<change_me>" \
-e MINIO_ACCESS_KEY="<change_me>" \
-e MINIO_SECRET_KEY="<change_me>" \
f1-recognition-model:v0.7
minikube image load f1-recognition-model:v0.7
bounding box with Team name and probability