Action Recognition with the Augmented MoCap Data using Neural Data Translation
The Augmented Data within the Florence3D-Action Dataset
Left to right, top to button. The original actions, the mirrored actions, the noisy actions (randomly added Gaussian noise into the original actions), mirrored action with Gaussian noise.
The Actions Generated by Our Approach
Left to right. The source actions (training), target actions (training), and the generated actions