Imagining the Unseen: Learning a Distribution over Incomplete Images with Dense Latent Trees
This work addresses a specific inference bottleneck in generative modeling for computer vision, offering an incremental improvement for tasks like image completion.
The paper tackled the problem of intractable inference in hierarchical generative models for images by proposing Dense Latent Trees (DLTs), which enable efficient exact inference and were applied to image completion on MNIST and Fashion-MNIST datasets, showing successful learning through latent state visualization.
Images are composed as a hierarchy of object parts. We use this insight to create a generative graphical model that defines a hierarchical distribution over image parts. Typically, this leads to intractable inference due to loops in the graph. We propose an alternative model structure, the Dense Latent Tree (DLT), which avoids loops and allows for efficient exact inference, while maintaining a dense connectivity between parts of the hierarchy. The usefulness of DLTs is shown for the example task of image completion on partially observed MNIST and Fashion-MNIST data. We verify having successfully learned a hierarchical model of images by visualising its latent states.