Porting this (https://github.com/facebookresearch/VideoPose3D) 3D human pose estimation model into Android for inference in the wild (on live video in real time). My goal is to use TfLite Posenet to get the 2D human keypoints and then feed them into the 3D model. Using PyTorch.
Are you able to successfully replace the detectron2 with posenet? Does it give a promising result? (I am trying to replace detectron2 with a speeder 2D estimation model)
Are you able to successfully replace the detectron2 with posenet? Does it give a promising result? (I am trying to replace detectron2 with a speeder 2D estimation model)