Contributions are welcome, but only Dlib shape-predictors models are accepted.
If you want to share your model, just submit a pull-request with:
- An explanation of what your model recognize, along with the accuracy of the model.
- A couple of example images (with detected points, bounding-box, etc).
- The steps to replicate your work (e.g. used dataset, training script, training parameters, etc).
Submitted models should be zipped: the preferred method is bzip2.
Every new model should fall into a category folder, so if it doesn't exists just create it and upload the model inside. For example, /face_landmarks is the category for models able to localize facial landmarks.