Depth prediction for nonrigid scenes from our multi-view depth estimator trained on WSVD.