This task has received a considerable amount of attention, particularly for classical ways of modeling distributions like structured prediction.

Inrealisticscenarios with very limited numbers of data samples, it can be challenging to identify a kernel powerful enough to distinguish complex distributions.

In this paper, we study the problem of unsupervised object segmentation from single images.

In experiments, VAEs trained with the new loss function generatedrealistic,high-qualityimagesamples.

However, much less effort has been devoted to exploring the adversarial robustness of pre-trained models.

Our novel setup brings new understandings to both disentanglement and equivariance .

Tothisendweproposeaflow-based image-to-image model, called FlowU-Net with Squeeze modules (FUNS), that allows us to disentangle the features while retaining the ability to generate highquality diverse images from condition images.

Neural Information Processing Systems http://nips.cc/