multi-view-unet for ACDC dataset Combining co-training with unet, using the autoencoder to split different sub dataset and produce pseudo label Compared with basline, our method decrease the Hausdorff distance and Volume error, currently working on decreasing the dice loss