In October I will be working on pose recognition signup/login using tensorflow.js (during my free’ time which means the time remaining from my free time which is already being used up to make a unity3d project). This means knowing what decisions to take very early on. The project is till now loosely defined as in “pose recognition” only implies that you can signup and login based on 1 or more poses which are your password. Keep in mind that this project is a pre-cursor to an eventual VR platform. Though it can easily be used by web platforms via a user’s webcam.
Also I’ve decided to take a completely project oriented approach to this. I will refrain from reading ALL the tutorials on tensorflow.js or machine learning which by itself is a huge topic. Since I am primarily concentrating on making a ML product that can be integrated into a web app.
- Should a set number of poses be provided to the user to choose from? For example 16 poses
2. A user can subsequently choose a sequence of 3 poses, which will be the “pose password” associated with their unique username
3. The user then confirms the poses by performing the sequence of poses again:
- Using tensorflow.js (which uses PoseNet) or some other library?
- How will I capture a user’s webcam stream while they do the poses? Will I record the stream and then validate if it’s the correct pose password at the backend?
These are the primary decisions that have to be taken right now in order for the project not to stagnate.