Deep Markov Random Field for Image Modeling
This work addresses a fundamental bottleneck in image modeling for computer vision researchers, offering a hybrid approach that enhances MRF capabilities.
The paper tackled the limited expressive power of Markov Random Fields (MRFs) in generative image modeling by proposing a novel MRF model using fully-connected neurons to capture complex pixel interactions, resulting in notable improvements over state-of-the-art methods on various low-level vision tasks.
Markov Random Fields (MRFs), a formulation widely used in generative image modeling, have long been plagued by the lack of expressive power. This issue is primarily due to the fact that conventional MRFs formulations tend to use simplistic factors to capture local patterns. In this paper, we move beyond such limitations, and propose a novel MRF model that uses fully-connected neurons to express the complex interactions among pixels. Through theoretical analysis, we reveal an inherent connection between this model and recurrent neural networks, and thereon derive an approximated feed-forward network that couples multiple RNNs along opposite directions. This formulation combines the expressive power of deep neural networks and the cyclic dependency structure of MRF in a unified model, bringing the modeling capability to a new level. The feed-forward approximation also allows it to be efficiently learned from data. Experimental results on a variety of low-level vision tasks show notable improvement over state-of-the-arts.