NP-DRAW: A Non-Parametric Structured Latent Variable Model for Image Generation
This work addresses the problem of generating images with better compositionality and interpretability for applications in low-data learning and latent space editing, representing an incremental advancement over existing structured models.
The paper tackles image generation by proposing NP-DRAW, a non-parametric structured latent variable model that sequentially draws image parts on a latent canvas, resulting in significant performance improvements over previous structured models like DRAW and AIR, with competitive results on datasets such as MNIST, Omniglot, CIFAR-10, and CelebA.
In this paper, we present a non-parametric structured latent variable model for image generation, called NP-DRAW, which sequentially draws on a latent canvas in a part-by-part fashion and then decodes the image from the canvas. Our key contributions are as follows. 1) We propose a non-parametric prior distribution over the appearance of image parts so that the latent variable ``what-to-draw'' per step becomes a categorical random variable. This improves the expressiveness and greatly eases the learning compared to Gaussians used in the literature. 2) We model the sequential dependency structure of parts via a Transformer, which is more powerful and easier to train compared to RNNs used in the literature. 3) We propose an effective heuristic parsing algorithm to pre-train the prior. Experiments on MNIST, Omniglot, CIFAR-10, and CelebA show that our method significantly outperforms previous structured image models like DRAW and AIR and is competitive to other generic generative models. Moreover, we show that our model's inherent compositionality and interpretability bring significant benefits in the low-data learning regime and latent space editing. Code is available at https://github.com/ZENGXH/NPDRAW.