1. transform : extract keypoints from files
2. prepare : build a train / validation dataset of keypoints files, should be run only once
3. display : you can check if open is working
4. train
5. test
6. infer

the transform could be included in the pipeline