Synthetic training data
Artificially generated examples (rendered images, simulated depth maps) used to train models when real labelled data is scarce or costly.
Synthetic training data is generated by simulation or rendering instead of manual photography. Labels can be exact because the simulator knows ground-truth depth, poses, or object masks.
For Kinect, Microsoft Research trained random forests on hundreds of thousands of rendered depth frames with known body-part labels, then deployed the model on real living rooms. Synthetic data is now common in robotics and supervised learning when collecting diverse real-world pose estimation sets is slow.