Fetching the paper…

FlowFormer++: Masked Cost Volume Autoencoding for Pretraining Optical Flow Estimation · Around