They appear to be using blender to generate data for machine learning! So if you want to be able to recognizes poses or hand gestures from video or image still

They appear to be using blender to generate data for machine learning! So if you want to be able to recognizes poses or hand gestures from video or image stills, you need data to train your network. The bottleneck is often high quality 'labelled' data - real data that requires people to laboriously label each frame and region with the 'ground truth'. However, if you automatically generate the poses using something like Blender, you can also automatically label, and also generate a huge variety of systematic variations in camera angles, lighting, levels of noise etc. Obviously there might be a worry that the Blender generated images may somehow inject their own bias, but they claim it's working really well. https://www.microsoft.com/en-us/research/uploads/prod/2019/09/2019-10-01-Synthetic-Data-with-Digital-Humans.pdf https://www.microsoft.com/en-us/research/uploads/prod/2019/0...