The scenario
A skeleton-based fall detection dataset is limited; a GAN augments synthetic skeleton sequences to improve classifier performance.
What is in this drawing
Read the decisions behind it.
01
Split dataset into train/test sets
02
Accept generated samples via discriminator realism check
03
Adjust GAN parameters if classifier performance does not improve
04
Deploy model after performance gain
Reusable when training data for rare fall events is scarce and synthetic motion generation is needed.
More like this