Few years ago, I worked on a ML/CV project using body poses to create an exercise classifier (basically identify and count reps of an exercise being done on video). I used a Google interface called Blaze Pose to pick up 32 different 3-dimensional body pose points to input into the classifier. Seeing these last couple of slides is gaming me realize that these transforms used is probably key for the interface to help pick up these poses, which is really cool since I never really explored that part of the project before.
Few years ago, I worked on a ML/CV project using body poses to create an exercise classifier (basically identify and count reps of an exercise being done on video). I used a Google interface called Blaze Pose to pick up 32 different 3-dimensional body pose points to input into the classifier. Seeing these last couple of slides is gaming me realize that these transforms used is probably key for the interface to help pick up these poses, which is really cool since I never really explored that part of the project before.