Adversarial learning network framework for 3D scene flow estimation. The FlowNet3D \cite{ref9} architecture is used as a generator to predict the scene flow at each point of \(PC_1\) and to obtain \(PC_2^*\). The discriminator generates the probability that \(PC_2\) is true and the probability that \(PC_2^*\) is true, from which the loss functions are designed to train the generator and the discriminator respectively. FC represents fully connected layer.